gh-123295: Update TSAN suppression comment and sample for set_default_allocator_unlocked

[Zheaoli]

Fix #123295

• Issue: free-threading: set_default_allocator_unlocked is also exits on Linux #123295

[colesbury]

You are correct that the race condition also occurs on Linux. It's not operating system specific.

I don't think it's worth changing the comment in the suppressions file.

Would you be interested in addressing the underlying data race? It's described in #111924.

Basically, there are some places where we have to use the default "raw" memory allocator for PyMem_RawMalloc and similar. In those cases it's not safe to use the possibly overridden allocator. Currently, we do this by calling _PyMem_SetDefaultAllocator(PYMEM_DOMAIN_RAW, ...) to temporarily restore the default raw memory allocator, but that pattern is not thread-safe because the default allocator is a shared, global variable.

We should add internal-only, non-overridable functions for the default raw memory allocator and use those instead of swapping the global allocator. For example, see _PyMem_DefaultRawMalloc from the nogil-3.12 fork.

[Zheaoli]

Would you be interested in addressing the underlying data race? It's described in #111924.

Great, I will try to make it done