feat(automation): run timeout, retry/backoff policy, and force-stop action

[cagdasyurekli]

Resolves #621

Summary

Implements run timeouts, exponential backoff retries for failed automation runs, and an explicit force-stop action so that hung tasks cannot silently hold the scheduler overlap guard indefinitely.

Key Changes

1. Per-task & Default Run Timeout (coworker/automation/models.py, coworker/automation/scheduler.py):
   • Added timeout_seconds: Optional[float] = 900.0 (default 15 minutes) to ScheduledTask.
   • Scheduler wraps run execution in asyncio.wait_for(...). When the timeout fires:
     • Run status is marked timed_out with descriptive error.
     • Scheduler unconditionally releases the overlap guard in finally:.
     • An on_timeout notification is parked in the user's InboxStore and broadcast via WebSocket toast events.
2. Error Retry & Exponential Backoff (coworker/automation/scheduler.py, coworker/automation/store.py):
   • Added max_retries: int = 0 and retry_backoff_seconds: float = 60.0 to ScheduledTask.
   • If a run fails with status == "error" and attempts remain (fresh.retry_count < fresh.max_retries):
     • Calculates backoff retry_backoff_seconds * (2 ** retry_count).
     • Advances retry_count += 1 and schedules next run at now + backoff without clobbering cron schedule.
     • Run trigger is marked as "retry".
   • Successful runs or user cancellations reset retry_count = 0 and return to the primary schedule.
3. Explicit Force-Stop (coworker/automation/scheduler.py, coworker/server/manager.py, coworker/server/app.py, surfaces/gui/):
   • Scheduler.force_stop(task_id) immediately discards _running_ids and cancels the in-flight asyncio.Task.
   • manager.force_stop_automation(task_id) halts running task engines and persists a cancelled run record.
   • Added POST /v1/automations/{task_id}/stop endpoint in FastAPI.
   • Added stopAutomation API function and UI "Force Stop" buttons in ScheduledView.tsx.
4. Comprehensive Unit Tests (tests/test_automation.py):
   • Verified timeout triggers, timeout notification callback, and overlap guard release allowing subsequent executions.
   • Verified exponential retry progression and retry exhaustion back into normal cron scheduling.
   • Verified force stop cancellation and POST /v1/automations/{task_id}/stop endpoint.