Add release_gil_before_calling_cpp_dtor annotation for class_

[rwgk]

Description

Closes #1446.

This PR is a backport of:

• Add release_gil_before_calling_cpp_dtor annotation for class_ google/pybind11clif#30088 (main PR)
• Fix minor oversight in pywrapcc#30088 (test code only). google/pybind11clif#30092 (minor fixes)

It introduces the py::release_gil_before_calling_cpp_dtor option (py::class_ argument). For background, see the description of google/pybind11clif#30088 and #1446.

• The core change (in pybind11.h class_::dealloc()) is to add ReleaseGIL-try-catch-ReaquireGIL-rethrow logic around the existing code calling the C++ destructor of wrapped objects.

• All the rest is more-or-less boilerplate code to add the py:: release_gil_before_calling_cpp_dtor annotation, then use it to enable the new feature if that annotation is provided by the user.

---

Note for completeness:

These are identical to the current versions on the pybind11clif main branch @ commit 4841661:

• test_class_release_gil_before_calling_cpp_dtor.cpp
• test_class_release_gil_before_calling_cpp_dtor.py

---

@rainwoodman (original reviewer) for visibility.

---

Suggested changelog entry:

A ``py::release_gil_before_calling_cpp_dtor`` option (for ``py::class_``) was added to resolve the long-standing issue #1446.

[rwgk]

@gentlegiantJGC could you please review (and formally approve if it looks good to you)?

(This PR is pretty much exactly as reviewed previously by @rainwoodman.)

[rwgk]

Looking at this with a fresh eye, I got suspicious about a potential data race in the test, and ChatGPT agrees that there is a problem:

• https://chatgpt.com/share/67b0edfc-0178-8008-9ae2-e321ac6178c3

It looks like a very easy fix. I'll add a commit in a minute.

[gentlegiantJGC]

The implementation looks fine to me.
I got a bit lost in the test code. You might want someone with more knowledge of the internals of pybind11 and python to look over this.

[rwgk]

The implementation looks fine to me. I got a bit lost in the test code. You might want someone with more knowledge of the internals of pybind11 and python to look over this.

Thanks @gentlegiantJGC!

Since the test was reviewed already (almost as-is), I'll go ahead with merging.