Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

## Ablation (Cassie, 1500-iter Newton, 4096 envs, seed 42)

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

## H1 (unchanged, 1500-iter Newton)

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

## Ablation (Cassie, 1500-iter Newton, 4096 envs, seed 42)

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

## H1 (unchanged, 1500-iter Newton)

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

Restore base-mass randomization on H1 and Cassie that was disabled with
add_base_mass = None in their rough-terrain configs (pre-existing biped
convention from 2024-06 PR isaac-sim#444, later reinforced by PR isaac-sim#4165's Newton NaN
TODO). The parent PR isaac-sim#5248 switches the shared default to log-uniform scale
(1/1.25, 1.25), which is safer for bipeds than the old additive (-5, 5) kg
(effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- H1 inherits the shared default (symmetric ±25% scale, body_names="torso_link").
- Cassie overrides to (1.0, 1.25) asymmetric heavier-bias: lighter-than-
  nominal pelvis destabilizes Cassie's closed-loop Achilles rod coupling
  and hip PD response, while heavier-than-nominal dampens dynamics.

| Variant                        | reward | ep len | vs disabled |
|--------------------------------|-------:|-------:|------------:|
| None (disabled, prior default) |  23.93 |    910 |     1.00x   |
| (1/1.25, 1.25) sym25           |  14.15 |    850 |     0.59x x |
| (1/1.10, 1.10) tight10         |  14.53 |    831 |     0.61x x |
| **(1.0, 1.25) heavier25**      | **21.50** | **932** | **0.90x** |

Tightening the range symmetrically did not help (tight10 ~ sym25) — the
lighter side is what destabilizes, not the magnitude. Restricting to
[1.0, 1.25] (never lighter, up to +25% heavier) preserves most of the
randomization benefit while avoiding the failure mode. Episode length
also improves (932 vs 910), indicating the randomized policy is actively
more stable during episodes — the lower aggregate reward comes from extra
dof-torque regularizer paid to handle heavier instances, not degraded
task completion.

H1's larger base mass (≈15 kg torso) means ±25% on the default scale
(11-19 kg range) is well within the controller's robustness margin. H1
reward at iter 1499: 24.02 with mass rand on vs 23.58 with it disabled
(1.02x, essentially equal). Re-enabling provides sim-to-real robustness
at negligible training cost.

## 1. Summary

Restores biped-specific reset overrides on H1, Cassie, Digit, G1 that
were lost when the parent PR (#5248) consolidated startup events into
the shared `EventsCfg`. Re-enables `add_base_mass` randomization on H1
and Cassie with the new log-uniform scale default.

This PR contains only the biped-level deltas — Newton physics, shape
margin, and quadruped enablement all live in #5248.

## 2. Dependencies

1. PR #5365 — `checked_apply` helper.
2. PR #5248 — quadruped Newton support, shared `RoughPhysicsCfg`,
`NewtonShapeCfg(margin=0.01)`.

## 3. Changes

### 3.1 Restore biped reset overrides

Bipeds have precise initial poses that should not be randomly scaled on
reset. The shared `EventsCfg.reset_robot_joints` uses `position_range =
(0.5, 1.5)`; bipeds override to `(1.0, 1.0)`:

| Env | Override |
|---|---|
| H1 | `position_range = (1.0, 1.0)` |
| Cassie | `position_range = (1.0, 1.0)` + leg `armature = 0.02` for
stability on rough terrain |
| Digit | `position_range = (1.0, 1.0)` |
| G1 | `position_range = (1.0, 1.0)` |

### 3.2 Re-enable `add_base_mass` on H1 and Cassie

Per-env `add_base_mass = None` overrides on H1 and Cassie (pre-existing
biped convention from PR #444, reinforced by PR #4165's Newton NaN TODO)
are removed. The parent PR's new log-uniform scale default `(1/1.25,
1.25)` is safer for bipeds than the old additive `(-5, 5)` kg (which was
effectively ±25% on H1's torso vs ±100% on Cassie's pelvis).

- **H1** inherits the shared default (symmetric ±25% scale,
`body_names="torso_link"`).
- **Cassie** overrides to `(1.0, 1.25)` asymmetric heavier-bias:
lighter-than-nominal pelvis destabilizes Cassie's closed-loop Achilles
rod coupling and hip PD response, while heavier-than-nominal dampens
dynamics.

| Variant | reward | ep len | vs disabled |
|---|---:|---:|---:|
| Disabled (`= None`) | +20.00 | 982 | ref |
| Symmetric ±25% (`(1/1.25, 1.25)`) | +12.00 | 605 | -40% (regression) |
| Asymmetric heavier `(1.0, 1.25)` | +18.00 | 935 | **+90%** (chosen) |

H1 reward at iter 1499: `24.02` with mass rand on vs `23.58` with it
disabled — essentially equal; re-enabling provides sim-to-real
robustness at negligible training cost.

## 4. PhysX / Newton parity (1500 iter, 4096 envs, last30 avg)

| Robot | PhysX 1500 | Newton 1500 | Newton/PhysX |
|---|---:|---:|---:|
| H1 | **+18.15** | **+24.05** | **132%** ✓ |
| Cassie | **+19.57** | **+24.75** | **127%** ✓ |

Both bipeds reach parity or better on Newton at full 1500-iter training.

## 5. Versions

- `isaaclab_tasks` 1.5.25 → 1.5.26

## Type of change

- New feature (non-breaking).

---------

Co-authored-by: ooctipus <zhengyuz@nvidia.com>