[gouyt13]

Summary

Reduce repeated test work while retaining distinct storage, routing, numerical and persistence coverage.

• Merge repeated HNSW add checks and remove redundant metric/initializer combinations: 15 fewer test instances with the relevant assertions retained.
• Run static mypy checks on source installs and the four Linux x86-64 wheel ABIs. Every wheel keeps installed declarations and runtime metadata checks; benchmark protocol tests remain in Python quality CI.
• Remove duplicate CI summary cases, unused test helpers and obsolete rotator fixture cleanup.
• Replace the GIL test's single-search scheduling assumption with bounded repeated calls and explicit worker readiness.

No library behavior, public API, index format or production dependency changes.

Testing

• python -m pytest tests/python tests/test_mypy.py: 793 passed, 6 skipped. Final GIL-only delta: all 30 concurrency tests passed.
• Portable core Release build with native optimization disabled: 129 passed. Final fixture-only delta: 23 affected tests passed.
• python -m unittest discover -s .github/scripts -p 'test_*.py': 48 passed; Pyright: 2 passed.
• Python/C++ formatting, Ruff, ShellCheck and git diff --check passed.
• Built and installed the current CPython 3.13 wheel; verified installed files against the wheel. cibuildwheel 4.2.0 resolved all 16 configurations with the intended test ownership.
• Independent read-only review passed. It reran GIL/mypy tests, rejected a no-GIL-release control, accepted a delayed worker, and checked all wheel configurations. The old GIL test's false failure was reproduced with a controlled worker delay.

The six skips cover four unavailable C++ example-checker cases and two foreign-platform wheel checks. Remote platform/ABI CI remains pending; full index C++ and sanitizer suites were not rerun for these test-only C++ changes. No speedup claim.