Record retried failures as flakyFailure elements in JUnit XML

[LouisDeconinck]

Summary

• pytest's junitxml plugin ignores reports with outcome rerun, so failures from retried attempts left no trace in --junitxml output
• A small per-config hook now stashes rerun reports and appends <flakyFailure> elements (and <flakyError> for setup/teardown failures) — the Maven Surefire / flaky plugin convention — to the test's final <testcase> when its last call report arrives
• No-op when --junitxml isn't used; works with xdist-forwarded reports since it keys on nodeid/worker_id/item_index/node

Closes #170

Test plan

• Test that eventually passes: final testcase has 2 flakyFailure elements with message + traceback, no <failure>
• Test that keeps failing: final testcase has flakyFailure(s) plus the real <failure>
• Full suite passes (252 tests)

[sylvesterkaczmarek]

I think setup failures can still disappear here. pending is only flushed when a later call report arrives, but if the final attempt also fails in setup there is no call report, so the earlier rerun setup errors never become <flakyError>. Could we flush on the terminal setup/teardown report too and add that case?

[LouisDeconinck]

Fixed in 8734762: pending is now flushed on any failed non-rerun setup/teardown report as well, so rerun setup errors become <flakyError> elements even when the final attempt fails in setup. Added a regression test.

[LouisDeconinck]

Fixed — pending rerun reports are flushed on the terminal setup/teardown report as well, so setup errors on earlier attempts always become <flakyError> elements, plus a regression test (8734762).