feat: subprocess-based worker tests with kill/restart scenarios

[PythonFZ]

Context

Current worker tests use threading.Thread with run_worker_loop — multiple ZnDraw clients in the same process, each polling jobs.listen(). This works well for claim isolation, lifecycle, and room scoping, but doesn't test true distributed scenarios.

Proposed

Add subprocess-based worker tests that exercise:

Worker kill/restart

• Worker picks up task → process killed mid-execution → task marked as failed (timeout/sweeper)
• Worker killed → new worker spawns → re-registers same jobs → picks up pending tasks
• Multiple workers killed simultaneously → remaining workers absorb workload

Server kill/restart

• Server dies while workers are connected → workers detect disconnect → reconnect when server returns
• Server restarts on same port (investigate: does SO_REUSEADDR / SO_REUSEPORT allow immediate rebind, or is there a TIME_WAIT issue?)
• Server restarts with fresh DB vs persistent DB — worker token invalidation behavior

Multi-worker subprocess patterns

• N workers as subprocesses, each with own ZnDraw client
• Submit M tasks → verify all M complete across N workers
• Kill worker subprocess → verify its claimed tasks get reassigned

Implementation considerations

• Use subprocess.Popen or multiprocessing.Process for workers
• Need reliable kill (SIGTERM/SIGKILL) and health-check patterns
• Server restart on same port: check if uvicorn's Server.shutdown() releases the socket immediately, or if a delay/retry is needed
• Consider pytest-timeout for tests involving process death

Related

• Current worker tests: tests/zndraw/worker/
• Worker resilience tests: tests/zndraw/worker/test_resilience.py (already tests SIO reconnect, but in-process)
• Server factory fixture: tests/zndraw/conftest.py (server_factory)