Add --max-module-reruns and --max-class-reruns scope caps

[LouisDeconinck]

Summary

• Adds --max-module-reruns N and --max-class-reruns N, capping the total number of reruns within each module / test class — extending the existing suite-wide --max-suite-reruns mechanism to the scopes requested in the issue
• Scope counters live in StatusDB (in-memory) and are shared through the socket StatusDB under pytest-xdist, so caps hold for parallel runs; slots are reserved atomically alongside the suite slot and released if a later scope is exhausted
• The teardown-time finalizer suspension check now uses the same combined cap check, so scope exhaustion doesn't leak suspended finalizers

Closes #213

Test plan

• Per-module cap limits reruns in one module but not others
• Per-class cap limits reruns in one class but not others; function tests unaffected by --max-class-reruns
• Negative values rejected; existing --max-suite-reruns tests still pass; full suite green

[sylvesterkaczmarek]

The class-scope key is derived with item.nodeid.rpartition('::')[0], but parametrized IDs may themselves contain ::. For example, test.py::TestC::test_x[a::b] produces a key ending in test_x[a, rather than the class node ID, so different parameter IDs can get separate counters and exceed --max-class-reruns. Could this derive the scope from the structured parent/class node instead of splitting the nodeid string, and add a parameter ID containing :: as a regression?

[LouisDeconinck]

Fixed in 2866443: module/class scope keys are now derived from item.listchain() (pytest.Module/pytest.Class nodeids) instead of splitting the nodeid, so parametrized IDs containing :: share one counter. Added a regression test with a::b-style param IDs.

[LouisDeconinck]

Fixed — class/module scopes are now derived from the node hierarchy instead of splitting the nodeid, so parameter IDs containing :: get a single counter, with a regression test for test_x[a::b] (2866443).