Fix subtest reruns with structured NodeId keys

[janmrow]

pytest-dev/pytest#14758 changed how pytest stores node IDs internally: instead of strings, it now uses NodeId objects.

Because of this, the private failed_subtests mapping uses NodeId objects as keys.

pytest-rerunfailures was still looking up failed subtests using string .nodeid values. So that failed subtests were not rerun with newer pytest versions.

This change uses .id when NodeId is available, and falls back to .nodeid for older pytest versions.

Tested locally:

• the full test suite passes with pytest 9.1.1 and pytest main at 431f3e1f5,
• subtest regression tests pass,
• related xdist/subtest tests pass.

Only the failed_subtests handling was changed. Other uses of .nodeid stay the same.