gh-159458: Fix free-threading data race in datetime strftime century normalization

[AnnaAr321]

In Modules/_datetimemodule.c, normalize_century() lazily initializes a function-local static int cache = -1; on the first strftime() call using %Y, %G, %F, or %C. Under free-threaded builds (Py_GIL_DISABLED), concurrent initial calls to strftime() race on reading and writing cache, triggering ThreadSanitizer data race warnings.
Use FT_ATOMIC_LOAD_INT_RELAXED and FT_ATOMIC_STORE_INT_RELAXED to atomically load and store cache in free-threaded builds with zero overhead in default GIL-enabled builds.

• Issue: Free-threading: data race on static cache in _datetime.normalize_century() #159458

[encukou]

Thank you.
AFAIK, this doesn't affect users in practice, so the NEWS blurb isn't really necessary. But since you already wrote it, we can keep it.

[miss-islington-app]

Thanks @AnnaAr321 for the PR, and @encukou for merging it 🌮🎉.. I'm working now to backport this PR to: 3.15.
🐍🍒⛏🤖

[miss-islington-app]

Thanks @AnnaAr321 for the PR, and @encukou for merging it 🌮🎉.. I'm working now to backport this PR to: 3.14.
🐍🍒⛏🤖

[bedevere-app]

GH-159536 is a backport of this pull request to the 3.15 branch.

[bedevere-app]

GH-159537 is a backport of this pull request to the 3.14 branch.

[StanFromIreland]

This was merged before I could review it, so a couple of notes for future PRs.

[StanFromIreland]

I think this test does not work in our current testing configuration, as it relies on the cache being empty, correct? In the CI job the whole suite runs in parallel threads, so by the time this test runs the cache is almost certainly already initialised...

[StanFromIreland]

Suggested change

	Fix a ThreadSanitizer data race on the static cache in ``normalize_century()`` of the datetime module under the free-threaded build.
	Fix a data race in :meth:`datetime.date.strftime` and :meth:`datetime.datetime.strftime` on the free-threaded build.

Please only include user-facing details in blurbs.