Use st_birthtime for file created timestamp on macOS/BSD

[ktaletsk]

Instrumental in implementing 'Created' column in JupyterLab and solving jupyter/notebook#7797

On macOS/BSD, st_birthtime provides the actual file creation time, while st_ctime (currently used) is inode change time that updates on chmod, chown, etc.

Changes:

• Add _get_created_timestamp() helper that uses st_birthtime when available and valid
• Fall back to st_ctime on Linux/Windows or when st_birthtime is invalid
• Add OverflowError to timestamp exception handling
• Add tests for timestamp behavior and edge cases

Platform	created source
macOS/BSD	st_birthtime
Windows	st_ctime (creation time)
Linux	st_ctime (inode change time)

---

AI Disclosure: Developed with Claude Code (Opus 4.5); code review via Cursor Agents (opus-4.5-thinking, gpt-5.2-high, gemini-3-pro). All code reviewed by author.

[krassowski]

I guess it fixes a bug of "created" being the same as "last_modified" on some OSes, so adding bug

[ktaletsk]

The CI failures are unrelated to the changes in this PR:

1. test_execution_state (failing on Ubuntu 3.9, Windows 3.9/3.11/3.12) - This is a known flaky test that intermittently fails with "Kernel did not start up in 60 seconds". The test is already marked with @pytest.mark.flaky.

2. Test Docs - Pre-existing Sphinx documentation warnings (forward reference resolution issues with sphinx_autodoc_typehints) that are treated as errors. These warnings exist on the main branch.

All tests related to the st_birthtime changes pass successfully.

I don't have permissions to restart the CI workflow - could a maintainer please re-run the failed jobs?

[krassowski]

I kicked the tests.

1. test_execution_state - I previously tried to fix this in Fix flaky test_execution_state test #1579 - help/other ideas welcome!
2. Test Docs - this is recent - I opened an issue to track the failing docs test Test Docs started failing on main #1595 - again, thoughts on how to resolve it welcome!