feat(run-tracking): store BullMQ jobId immediately after run creation

[aaight]

Summary

Store the BullMQ jobId on the agent_runs record immediately after run creation in the worker container. This enables the cancel flow mapping (runId → jobId → container) needed for the cancel listener implementation.

Implementation Details

• Modified tryCreateRun() in src/agents/shared/runTracking.ts to call updateRunJobId(runId, process.env.JOB_ID) after successful run creation
• The JOB_ID env var is already passed to worker containers by the Router's buildWorkerEnv() in container-manager.ts
• Job ID storage is wrapped in a try/catch — failure is logged but does not block agent execution (fire-and-forget pattern)
• updateRunJobId is NOT called when JOB_ID env var is absent (local dev / non-Docker mode)

Testing

Added comprehensive unit tests to verify:

• updateRunJobId is called with correct runId and jobId when JOB_ID is set
• updateRunJobId is NOT called when JOB_ID is unset
• Failure to store jobId is logged but does not block run creation
• All existing tests pass with no regressions

Verification

• Type checking: ✅ PASS
• Linting: ✅ PASS (no fixes applied)
• Tests: ✅ PASS (12 tests in runTracking.test.ts)
• All agent tests: ✅ PASS (569 tests)

Closes: https://trello.com/c/afFdwvgk/304-as-a-developer-i-want-the-worker-to-store-its-bullmq-jobid-on-the-run-record-so-that-cancel-commands-can-map-runs-to-containers

[nhopeatall]

Summary

LGTM — Clean, well-scoped change that stores the BullMQ jobId on the run record immediately after creation, enabling the cancel flow mapping. The implementation correctly follows the existing fire-and-forget error handling pattern used throughout runTracking.ts.

Verified:

• updateRunJobId exists in runsRepository.ts and performs a straightforward DB update on the jobId column (confirmed in schema at src/db/schema/runs.ts:36)
• JOB_ID env var is confirmed set by buildWorkerEnv() in container-manager.ts:81
• Inner try/catch correctly isolates jobId storage failure from run creation — the run ID is still returned even if updateRunJobId throws
• Guard on process.env.JOB_ID correctly skips the update in local dev / non-Docker mode
• Tests cover all three paths: happy path, absent env var, and failure resilience
• All CI checks pass