panic on attach if interpreter is shutting down

[davidhewitt]

This is an attempt to solve the build failure seen in #5309 (at least locally, it seems to be good enough).

The problem is that the test added in #4874 which repeatedly calls Python::attach to try to hit the "hang thread" case seems to just cause Python 3.14 to segfault. The docs explicitly say this might happen if you call PyGILState_Ensure during finalization, to be fair.

To try to avoid that crash, this adds the FFI definition Py_IsFinalizing (available on 3.13 and up) and uses that to try a best-effort panic instead of segfault. As a follow-up I'll probably expose Python::try_attach and change the test to use it, to avoid a slightly messy panic at the end of the test suite.

While adding Py_IsFinalizing I synced the pylifecycle definitions and corrected a few other cases.

[codspeed-hq]

CodSpeed Performance Report

Merging #5317 will not alter performance

_{Comparing davidhewitt:attach-panic-shutting-down (d4febfe) with main (12a1792)}

Summary

✅ 93 untouched benchmarks

[davidhewitt]

Thanks for the review! FYI I'm just drafting a follow up to expose Python::try_attach.