Fix race condition when closing jitdump file in perf JIT trampoline

[rgfegegeegege]

The perf JIT trampoline code writes to a shared FILE* from multiple threads
(perf_map_jit_write_entry can be called concurrently when generating JIT code),
but it does so without holding the map_lock.

During interpreter shutdown, perf_map_jit_fini() closes the file while holding
the lock, but sets perf_map to NULL only after releasing the lock.

This opens a small race window where a concurrent write could:

• read a stale perf_map pointer after fclose(), or
• call fwrite() on a closed (or NULL) FILE*, leading to undefined behavior
  (likely a crash or corrupted jitdump file).

Although the JIT perf support is experimental and the race is rare,
it is a real thread-safety bug.

This change fixes it by:

• Acquiring the map_lock around all writes in perf_map_jit_write_entry().
• Adding a double-check after acquiring the lock (in case fini() ran while
  waiting for the lock).
• Moving the perf_map = NULL assignment inside the locked section in fini()
  for consistency.

No functional changes in the single-threaded case, and no regression in
jitdump file format or perf compatibility.

Fixes a potential use-after-close (CWE-910).

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[python-cla-bot]

The following commit authors need to sign the Contributor License Agreement:

• [email protected]