[cschanhniem]

Summary

• add a machine-readable benchmark metrics snapshot builder and CLI entrypoint
• add doc consistency coverage that checks the generated snapshot against authoritative current metrics
• correct stale expert-ablation numbers in roadmap and benchmark docs

Verification

• make metrics-snapshot
• uv run pytest -q tests/test_metrics_snapshot.py tests/test_docs_consistent.py tests/test_cli.py tests/test_expert_ablation.py -q
• pre-push hook: .venv/bin/python -m pytest -q (1498 passed, 7 skipped)