Add --xfail-flaky to report exhausted flaky failures as xfail

[LouisDeconinck]

Summary

• Adds an opt-in --xfail-flaky flag: a test that still fails after all its reruns are exhausted is reported as xfailed instead of failed (functionality similar to pytest-ignore-flaky, as requested)
• Only applies to tests that were actually rerun (execution_count > 1); tests failing on first attempt or excluded by --only-rerun/--rerun-except still report as failed
• Implemented by converting the final call report to skipped + wasxfail, so pytest's standard xfail reporting applies

Closes #92

Test plan

• Always-failing test with --reruns 2 --xfail-flaky → 1 xfailed, 2 rerun, exit code 0
• Eventual pass unaffected; tests without reruns unaffected; marker-based reruns covered
• --only-rerun non-matching error still reports failed

[sylvesterkaczmarek]

--xfail-flaky only converts the final report when report.when == 'call'. Reruns can also be triggered by setup or teardown failures, so a fixture that fails on every attempt can exhaust its reruns and still finish as a normal failure even though the option says exhausted flaky failures are reported as xfailed. Could the terminal setup/teardown failure be handled too, or the option be explicitly narrowed, with a setup-failure regression test?

[LouisDeconinck]

Fixed in 53b61ec: the xfail conversion now applies to the terminal failed report in any phase, so a fixture that fails on every attempt ends as xfailed. Added a setup-failure regression test.

[LouisDeconinck]

Fixed — exhausted setup/teardown failures are now converted to xfail too, not just call reports, with a setup-failure regression test (53b61ec).