feat: unified management stop event

[tomerqodo]

Benchmark PR from qodo-benchmark#171

---

Note

Centralizes execution stop control via a shared threading.Event managed by GraphEngine, improving coordinated shutdown and cancellation.

• Engine/runtime: Create shared stop_event in GraphEngine; clear on start and set on stop; attach to GraphRuntimeState and pass to subsystems
• Dispatcher: Use shared stop_event (remove internal one); reduce join timeout to 2s
• WorkerPool/Worker: Pass shared stop_event to Worker; make Worker.stop() a no-op; pool stops by joining workers (2s) without per-worker stop signaling
• Nodes: Add _should_stop() in base Node and check between yields; emit NodeRunFailedEvent with "Execution cancelled" on stop
• Runtime state: Add stop_event field to GraphRuntimeState
• Tests: New test_stop_event.py and updates to dispatcher tests validating propagation, resume behavior, and shorter timeouts

^{Written by Cursor Bugbot for commit ef61c4b. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Worker scale-down fails because stop() is now no-op

Medium Severity

The _remove_worker method calls worker.stop() expecting it to stop an individual worker during scale-down operations. However, Worker.stop() is now a no-op that does nothing since workers now rely on a shared stop_event from GraphEngine. This means _try_scale_down will fail to actually stop workers — the join(timeout=2.0) will always timeout and the worker thread continues running in the background, defeating dynamic scaling and causing resource waste.

Additional Locations (1)

• api/core/workflow/graph_engine/worker.py#L72-L78