add test coverage for weird case of plugin with multiple impls of same spec

[wahuneke]

These tests are attempting to find @RonnyPfannschmidt 's theoretical issue described in issue #431 . Conversation in 431 is ongoing. This PR is here for reference. It is a DRAFT.

NOTE: these tests currently pass. They do not detect any new issues.

[RonnyPfannschmidt]

well done, those tests verify that my assessment from just/only looking at the code was wrong and now we have a demonstration of the correctness for those cases

[RonnyPfannschmidt]

Rebased onto current main and updated for merge:

• Dropped the test_hook_multi_impl commit — the behaviour it tested (register/unregister with multi-impl plugins, get_hookcallers return count) is now fully covered by test_unregister_plugin_with_multi_hookimpls and test_get_hookcallers_no_duplicates added in Fix _remove_plugin and get_hookcallers for multi-impl plugins #646, which also fixed the underlying bug (multi hook registration does not unregister #431) that this PR was originally investigating.

• Kept the tracer tests (test_plugin_tracing, test_dbl_plugin_tracing) — these are new coverage for hook tracing with multi-impl plugins, not covered elsewhere.

• Fixed List[str] → list[str] and added a missing blank line to pass pre-commit (ruff, flake8, mypy).