Avoid TSan warnings in faulthandler

[colesbury]

Bug report

The faulthandler module's code for printing stack traces is not thread-safe:

• That's mostly okay -- faulthandler is mostly triggered when something goes wrong, like a segfault or a deadlock/timeout, so best effort is acceptable.
• In free threaded Python, we don't print out stacktraces for all threads on a segfault, but we do print them on a timeout. That can lead to TSan reported data races on timeouts.

I think we want to keep the timeout behavior the same, which means living with some data races:

• It mostly works and is useful.
• There's no one "current thread" in a timeout. The "dump traceback later" thread (faulthandler_thread) doesn't have a thread state / hold the GIL, so it's a different situation than on segfault, where the crashing thread's stacktrace is the most useful (and safest to access).

We also have some suppressions in suppressions_free_threading.txt, but these are out of date.

I think we should add _Py_NO_SANITIZE_THREAD to the relevant faulthandler.c functions so that TSan doesn't report data races when faulthandler is running.

Linked PRs

• gh-143108: Don't instrument faulthandler.c for TSan #143109