Do not infer Python runners from stub-only tests

[hzw0813]

Summary

• require an executable .py test file before layout-based Python framework discovery can infer pytest or unittest
• keep .pyi files available to static Python analysis and test-like relationship discovery
• add regressions proving stub-only repositories do not invent runtime checks while stub relationships remain visible and stop at runner qualification
• document the static-vs-runnable distinction and regenerate committed dist

Why

detectPythonTests() accepted both .py and .pyi test-shaped filenames even though Python target qualification already rejects .pyi. That meant a repository containing only test-shaped type stubs could cause --run-checks to launch python -m pytest or python -m unittest despite having no executable Python test file. This crossed ProofDiff's own separation between broad test-like relationship discovery and runnable-target identity.

The fix narrows only framework discovery. Stub files remain valid source/static relationship inputs; they simply cannot create an executable runner by themselves.

Verification

• full npm test passed after the change
• clean npm run build regenerated dist
• dedicated regressions cover both pytest-like and unittest-like .pyi stubs and static relationship preservation
• temporary updater files were removed from the final diff