test_intl: resolve test flakiness by using integer nanosecond timestamps during file modification detection

[jayaddison]

Feature or Bugfix

• Bugfix

Purpose

• The intent here is to improve the reliability of the tests within the test_intl module, and in particular to resolve Seemingly random CI failures of tests/test_intl.py::test_gettext_dont_rebuild_mo with Windows #11232.

Detail

• The flaky test assertions relate to the detection of modified files during application builds; and it seems to be message catalog files (.mo) that are relevant.
• Timestamp detection had been using floating-point seconds-since-the-epoch.
• Based on a small sample size of testing from Testing: pr 11392 in combination with: pr 11426, pr 11432, and a revert of bug 440 jayaddison/sphinx#3, I believe that integer-based nanoseconds-since-the-epoch may be more reliable (but I'm not 100% confident).
• At the same time, revert commit 718993a that I believe should be unnecessary -- and maybe-cause-related -- simplifying the code and retrieving the current-time as nanoseconds-since-epoch.

Relates

• May resolve Seemingly random CI failures of tests/test_intl.py::test_gettext_dont_rebuild_mo with Windows #11232.

[AA-Turner]

Thanks @jayaddison!

A

[AA-Turner]

https://github.com/sphinx-doc/sphinx/actions/runs/5615685910/job/15216466568

A test failure on test_gettext_dont_rebuild_mo...

FAILED tests/test_intl.py::test_gettext_dont_rebuild_mo - AssertionError: assert 1 == 0
 +  where 1 = <function test_gettext_dont_rebuild_mo.<locals>.get_number_of_update_targets at 0x0000020DF2DD83C0>(<SphinxTestApp buildername='dummy'>)
= 1 failed, 1821 passed, 23 skipped, 5 xfailed, 28 xpassed in 212.44s (0:03:32) =