Fix last-failed cache for unittest subtest failures

[Kerim-Sabic]

Failed unittest.TestCase.subTest checks can disappear from the last-failed cache when a later subtest or the parent's call report passes. Consequently, --last-failed --lfnf=none can deselect a test that still fails.

Keep failures from SubtestReport until the parent call report has been processed. Consume the per-execution marker so a subsequent successful execution of the same node ID still clears its cached failure.

Closes #14575.

The regression tests cover passing and skipped reports after a failure, errors inside subtests, a skipped parent, operation without terminal reporting, rerunning with --last-failed, clearing the cache after repair, and repeated execution of one node ID.

Validation on Windows / Python 3.13.5:

• tox -e linting: passed.
• Final changed-file pre-commit hooks and git diff --cached --check: passed.
• tox -e py313 -- -n 4 --basetemp=../pytest-test-tmp/full-py313-final: 4,615 passed, 145 skipped, 13 xfailed, 1 xpassed, and 1 failed. All 11 added regression cases passed.
• The only failing test, test_fixture_doctest_skip_has_line_number, also fails at unchanged base commit aec0291ed14c724873bfcea99d5dfd739f75a68f in this environment.
• Against unchanged production code, the final regression tests produce 10 failures from missing cache entries and one passing duplicate-execution control.

AI assistance: Codex helped investigate the report sequence, implement the change, and run validation. Kerim Sabic confirmed an in-depth review of the prepared patch and explanation before submission and is the human contributor responsible for the change.

• Added regression tests.
• Added changelog/14575.bugfix.rst.
• Added Kerim Sabic to AUTHORS in alphabetical order.
• Allow maintainers to push and squash commits.

No feature documentation is needed for this bugfix. AI assistance is disclosed above; optional AI coauthor trailers have not been added.