gh-126016: Fix flaky test by allowing the SIGINT return code

[ZeroIntensity]

In GH-139158, I added a test that sent KeyboardInterrupt to the main thread so it would interrupt _Py_Finalize and shut down. There was a subtle race condition there:

1. The main thread starts the subthread by calling _thread._start_joinable_thread.
2. The thread is physically created before _start_joinable_thread finishes from Python's perspective, so it begins execution.
3. The subthread prints out a, signaling to the parent process that it is ready and may send the interrupt signal.
4. The subthread yields the GIL anywhere after this point.
5. The main thread acquires the GIL, and then finishes executing _start_joinable_thread.
6. The parent process sends the KeyboardInterrupt before _Py_Finalize is reached. Python returns -2, which fails the test.

The simplest way to fix this is by allowing a return code of -2 (-SIGINT) in addition to 0.

• Issue: Multithreaded subinterpreters can be running during finalization #126016

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @ZeroIntensity for commit 2a28762 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F139219%2Fmerge

If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again.

[encukou]

The buildbot failures are unrelated, pre-existing issues.

[encukou]

Thank you!