- Add CronWorkflowTrigger type with schedule, timezone, overlap_policy, timeout
- Add WorkflowRun type for execution history tracking
- Add workflow_runs DB table with indexes
- Validate cron triggers in workflow validation (exactly one trigger source)
- Support overlap policies: skip, queue, parallel

- Fix workflow_normalize.go syntax error (extra closing brace)
- Fix cron_scheduler.go to use pointer iteration over ws.Workflows
- Fix cron_scheduler.go to use cs.cron.Schedule() correctly (returns EntryID only)
- Remove unused context import from cron_scheduler.go
- Remove unused types import from cron_api.go
- Wire cron scheduler start/stop into server lifecycle

- go mod tidy promoted robfig/cron from indirect to direct
- Removed stale testcontainers dependencies no longer used in codebase

- Fix RunContext scan: scan into string then json.Unmarshal into map[string]interface{}
- Add finishRunByClawID to mark runs completed/failed/canceled by claw ID
- Wire finishRunByClawID into all claw terminal status paths:
  * idle (handleClawDoneSignal) -> completed
  * manual kill (server.go) -> canceled
  * error (stopAgentWithReason) -> failed
  * issue left trigger (github_issues.go, linear.go) -> canceled
  * PR merged (pr_watcher.go) -> completed
  * pipeline terminal (pipeline_runner.go) -> completed
  * factory trigger failures (github_issues, linear, shortcut, external) -> failed
  * factory trigger reclaimed (factory_triggers.go) -> canceled

- recordRun now accepts runID as first parameter instead of generating a new one
- Fixes updateRun always matching zero rows because runID was never in DB
- Fixes error path leaving 'running' row stuck forever while inserting new 'failed' row
- Skipped runs still generate their own runID (no updateRun needed for them)

- Add cronTriggerNotFoundError type for 'not found' / 'not cron-triggered' cases
- manualTrigger returns typed error for 404 cases, plain error for DB failures
- API handler returns 404 for not-found, 500 for infrastructure errors

- Change running map from bool to int (counter)
- Increment counter when run starts, decrement when it finishes
- Check > 0 instead of true for active runs
- Prevents parallel run defer from corrupting tracker when another run is still active

Addresses Greptile review comment: the error path was calling recordRun
again with the same runID, which caused a primary key conflict and left
the run stuck in 'running' status forever.

Now uses failRun() which UPDATEs the existing row with status='failed',
result message, and finished_at timestamp.

The running counter was decremented immediately after claw creation,
making skip/queue overlap policies ineffective since the claw was still
running when the next cron firing occurred.

Now the counter stays elevated until finishRunByClawID is called when the
claw reaches a terminal status (idle, deleted, error). A clawWorkflow map
tracks clawID -> workflow key for this purpose.

Addresses Greptile review comment about overlap enforcement being a no-op.

…olicy

runWorkflow now returns a bool indicating whether a run was actually started.
When overlap policy causes a skip, manualTrigger returns a cronTriggerSkippedError
which the API handler translates to HTTP 409 Conflict instead of 200 OK.

Addresses Greptile review comment about manualTrigger silently returning success
when the run is skipped.