Cover broken job command paths with integration tests

[CoreyRDean]

Non-technical summary

This closes the remaining test gap around clauck's broken-job recovery commands. clauck broken, clauck revive, and clauck discard now have direct command-level coverage, so future changes to the tombstone flow are less likely to regress silently.

Why this matters now: issue #124 is still open on main, and the repo has already needed follow-up fixes in these revive paths. The helper-level tests were no longer enough proof for the user-facing command surface.

The system is better afterward because the recovery flow now has a stronger safety net across happy paths, DAG resume behavior, and operator-facing failures.

Technical summary

• Expanded tests/test_clauck.py to cover:
  • cmd_broken empty output
  • cmd_broken table output with stale markers and stable ordering proof
  • cmd_discard removal counts and missing-tombstone behavior
• Expanded tests/test_clauck_revive.py to cover:
  • DAG resume dispatch through dag-runner.py --resume
  • revive override contents on the DAG path
  • missing tombstone, stale tombstone, and missing-session-id error exits
• Kept the increment test-only: no runtime behavior changes, only stronger verification around the existing command contract.

Relevant intent: this advances clauck's inspectability and trust-via-proof contract by ensuring the user-facing recovery commands have observable, durable regression coverage.

Tests:

• python3 -m unittest -v tests.test_clauck tests.test_clauck_revive
• python3 -m unittest discover tests -v

Breaking changes: none.

Additional notes

Trade-off: this does not add end-to-end subprocess execution for revive; it stays at the command boundary with subprocess.Popen mocked so the suite remains fast and isolated.

Deferred follow-up: issue #124 mentioned broader integration coverage for these command flows. This PR covers the command branches directly, but does not add real spawned-job fixtures beyond the existing run-job.sh resume-precedence test.

Remaining gap: if the repo later wants deeper end-to-end proof for revive side effects, that should be a separate increment rather than broadening this bounded test-only change.