fix: handle race condition in daily warning v0.x

[player3zxcvbnm]

Hi

I’ve implemented a fix for the race condition in _warn_once_per_day as discussed in #2530.

The Problem:
On Windows, parallel execution (e.g., pytest -n) causes a FileNotFoundError. This is due to multiple processes clashing over the temporary file creation and renaming process.

The Fix:

• Replaced the "temporary file + rename" logic with an atomic lock-file pattern.
• Uses open(path, "x") to ensure only one process creates the stamp file.
• Catches FileExistsError so secondary processes skip the update rather than crashing.
• Preserves the parents=True directory creation from Create parent directories for cache #2525.

Testing:
Verified on Windows 11. Parallel tests that previously failed now pass consistently.

Closes #2530

[player3zxcvbnm]

Changes:

Encoding: Added encoding='utf-8' to file reads/writes.
Linting: Cleaned up unused imports and whitespace.

CI Failures:
They are all TypeErrors in the mcse diagnostics (specifically the sd method). These appear to be pre-existing environment issues.

Let me know if there's anything else I should adjust on my end.

[OriolAbril]

@Zethson or @schmoelder would either of you be able to test this PR? If it works I can make a patch release.

The version comment will need fixing before merging, but the other two can be ignored if the PR already fixes the issue.

[OriolAbril]

undo

[OriolAbril]

I am not sure accumulating these lockfiles over time is a very good idea. We don't want the warning to be too overbearing but if it shows multiple times when importing arviz in parallel it would be perfectly fine too. If it works it might be fine though, this is a backport fix on a frozen version so no need to overthink it too much

[player3zxcvbnm]

Applied the suggestion for the lockfile and bumped the version to 0.23.4. Ready for testing! My apologies for the repo confusion earlier—I'm still getting oriented with the ArviZ ecosystem.