Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@CHANGELOG.md`:
- Line 30: Update the Changelog entry for deprecating SensorContact.shape and
SensorContact.reading_indices to include explicit migration guidance: state
which new field(s) replace each deprecated property (referencing
SensorContact.shape and SensorContact.reading_indices), provide a short example
or pseudo-code showing how to map old values to the new fields, and call out any
serialization or API compatibility steps users must take (e.g., updating stored
records or client code). Ensure the text is concise and actionable so users can
follow a clear migration path.

---

Nitpick comments:
In `@newton/_src/sensors/sensor_contact.py`:
- Around line 155-166: The _ensure_sorted_unique function currently uses a
recursive sort when it encounters an out-of-order pair; replace that with a
simpler always-sort approach: sort the indices up front (use sorted(indices)),
then iterate once to detect duplicates and raise ValueError with the same
message using param_name and the duplicate value, and finally return the sorted
list; update the body of _ensure_sorted_unique to remove recursion and always
return the sorted, duplicate-checked indices.

                if attr.startswith("time_"):
                    # Warmup run (not measured).
                    method(*params)
                    # Run timing benchmarks multiple times and measure elapsed time.
                    for _ in range(number):
                        start = time.perf_counter()
                        method(*params)
                        t = time.perf_counter() - start

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@newton/_src/sensors/sensor_contact.py`:
- Around line 495-496: The metadata sets counterpart_type to "body"/"shape"
unconditionally based on counterpart_is_body even when counterpart_indices is
empty or force_matrix is None, making it inconsistent; update the assignment for
counterpart_type (where sensing_obj_type and counterpart_type are set) to
produce None when there are zero-column configs—i.e., only set "shape" or "body"
if counterpart_indices contains at least one non-empty entry (or force_matrix is
not None/has columns); otherwise set counterpart_type = None so it stays in sync
with counterpart_indices and force_matrix.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@newton/_src/sensors/sensor_contact.py`:
- Around line 445-446: The current allocation of counterpart columns uses
max_readings computed before filtering to worlds that actually have sensing
rows, causing unused columns (see _assign_counterpart_columns,
counterpart_world_start, world_count, max_readings, counterpart_indices). Update
the logic so max_readings is derived only from counterparts whose worlds
intersect the sensing rows' worlds: first compute the set/list of worlds that
have sensing rows from counterpart_indices (or reject counterparts mapped to
worlds with no sensing rows), then call or modify _assign_counterpart_columns
(or its caller) to calculate max_readings and build
counterparts_by_world/_net_force based on that filtered set so columns are only
allocated for worlds that expose sensing rows.

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@asv/benchmarks/simulation/bench_sensor_contact.py`:
- Around line 57-58: Class-level benchmark parameter attributes use mutable
lists (params and param_names) which violates RUF012 and risks accidental
mutation; change them to immutable tuples by replacing ["ant", "humanoid"] and
[64, 8192] with ("ant", "humanoid") and (64, 8192), and replace ["robot",
"world_count"] with ("robot", "world_count"); also audit and convert any other
class-scope parameter lists in this file (the other occurrences of
params/param_names at the same class scope) to tuples to ensure immutability.