Ninth PR in the 12-PR file-policy rollout. Adds `cargo xtask
policy-report` which runs every advisory check and aggregates the
results into target/policy/policy-report.{md,json}.

## Behavior

policy-report invokes each of the seven check functions in advisory
mode:

  check_file_policy::check
  checks::check_generated
  checks::check_executable_files
  checks::check_dependency_surfaces
  workflow_checks::check_workflow_surfaces
  workflow_checks::check_process_policy
  workflow_checks::check_network_policy

Each sub-check writes its own target/policy/*-report.json (as in
previous PRs). policy-report then reads back the seven JSON artifacts,
lifts each one's `summary` block, picks a headline metric per area
(unreceipted | invalid_policy_refs | unknown_total | missing_fields |
expired | stale | unused — first non-zero wins; falls back to a
universe-size-keyed "clean" row), and emits a unified MD + JSON.

## Decisions

- **Advisory only.** No --mode flag. Promotion to blocking is the job
  of PRs 10/11/12; they tighten CI's invocation of the individual
  checks. Adding modes here would duplicate state.

- **Re-run vs read-only.** policy-report always re-runs the sub-checks
  so the aggregated report reflects current state, not stale artifacts
  from a prior session. Cost is small — each sub-check is a few
  hundred ms.

- **Untyped JSON re-read for aggregation.** Each sub-report has a
  different schema. Lifting the `summary` block as `serde_json::Value`
  is simpler and more robust than maintaining seven typed deserializers
  in xtask. The headline metric is computed from the same Value.

## Acceptance

- cargo check --workspace --locked passes.
- cargo clippy -p xtask --all-targets --locked -- -D warnings clean.
- cargo fmt --all -- --check clean.
- `cargo xtask policy-report` ran on this branch:
    wrote unified policy-report (7 areas, 7 headline rows)
- target/policy/policy-report.md and target/policy/policy-report.json
  both produced.

Headline on current main:

  | Non-Rust file policy | unreceipted    | 945 |
  | Generated files      | _clean_        | 0 (universe) |
  | Executable files     | _clean_        | 0 (universe) |
  | Dependency surfaces  | _clean_        | 16 (universe) |
  | Workflow surfaces    | _clean_        | 9 (universe) |
  | Process policy       | unknown_total  | 10 |
  | Network policy       | _clean_        | 9 (universe) |

The two non-clean rows match the documented expected state: 945
unreceipted files because PR 2's seeded receipts intentionally cover
only a slice of the workspace, and 10 false-positive process tokens
because the PR 8 detector is grep-style (see #220 PR body OPEN
question).

## Out of scope

- PR 10 (#209) — wire the advisory checks into CI as a job and upload
  target/policy/ as an artifact.
- PR 11 (#210) — promote file/generated/executable/dependency/workflow
  checks to blocking.
- PR 12 (#211) — promote process/network checks to blocking.

Closes #208.