Fix: Timing-related issue with clean-shutdown tests.

[griffinmilsap]

Summary

This PR fixes a timing race in shutdown test orchestration that could cause inconsistent return codes in test_shutdown_exception_on_cancel.

Problem

tests/shutdown_runner.py emitted READY as soon as runner.running became True.
In some runs, that happened before the target task actually entered its execution loop.

As a result:

• SIGINT could be sent too early
• cancellation-path behavior under test might not be exercised
• subprocess could exit 0 instead of a SIGINT code (130 / -SIGINT)

Changes

Updated tests/shutdown_runner.py to synchronize readiness with actual task start:

• Added a module-scoped STARTED event.
• Set STARTED at entry of each task used by shutdown tests:
  • BlockingDiskIO.blocked_read
  • BlockingSocket.blocked_recv
  • ExplodeOnCancel.explode
  • StubbornTask.ignore_cancel
• Changed READY gate in watcher thread from:
  • runner.running
    to:
  • runner.running and STARTED.is_set()

Why this fix

This ensures signal injection only starts after the target task is truly active, making shutdown behavior tests deterministic and aligned with the scenario they are intended to validate.

Testing

• pytest tests/test_clean_shutdown.py::test_shutdown_exception_on_cancel -q
• pytest tests/test_clean_shutdown.py -q

Both pass consistently with this change.