Slice 242: Runtime Recovery and Watchdog Sweep

[fourtytwo42]

Canonical slice for runtime recovery/replay/watchdog hardening after Slice 241. Scope: runtime-only recovery correctness for pending flows, prompts, outboxes, and resume behavior; no API/schema changes.

[fourtytwo42]

Slice 242 complete.

Commit: 08d5065

Validation:

• python3 -m unittest -v apps/agent-runtime/tests/test_runtime_services.py apps/agent-runtime/tests/test_trade_path.py apps/agent-runtime/tests/test_approvals_run_loop.py apps/agent-runtime/tests/test_liquidity_cli.py apps/agent-runtime/tests/test_runtime_invariants.py -> Ran 234 tests, OK
• npm run db:parity -> ok: true, checkedAt: 2026-03-08T21:29:41.667Z
• npm run seed:reset -> ok: true
• npm run seed:load -> ok: true, loadedAt: 2026-03-08T21:29:49.748Z
• npm run seed:verify -> ok: true
• restored tracked infrastructure/seed-data/.seed-state.json after seed commands
• npm run build -> passed
• pm2 restart all -> xclaw-web online

Recovery coverage added:

• corrupted pending intent/spot-flow files can be rewritten after restart
• mirror failure preserves executing transfer flow across restart
• persisted approval prompt survives restart and resend cooldown
• trade-usage replay survives restart with deduplicated remaining queue
• stale transfer resume converges once across restart
• limit-order run-once replays queued status after restart
• liquidity resume reuses execute contract for approved intent