feat: add safe pytest and unittest migration

[polishdataengineer]

Summary

• add testenix migrate {auto,pytest,unittest} with dry-run, validation-only, and safe publication modes
• convert the supported static pytest subset and wrap supported unittest suites without changing original files
• validate source and generated suites in isolated shadow copies before an atomic create-only publish
• compare exact per-test and aggregate outcomes, re-hash the source before publication, and emit a machine-readable migration report
• document supported/blocked constructs, safety guarantees, recovery behavior, CLI examples, and LLM-ready references
• add reproducible 3,000-test migration benchmarks and expose them in README and GitHub Pages
• test migration from built wheel and sdist artifacts, including unittest execution from an unrelated working directory

Transaction and safety guarantees

• originals are never overwritten, moved, or deleted
• output and report paths must be new, disjoint from the source, and inside the project
• validation failure leaves the original suite intact and publishes no output
• publication is descriptor-anchored and atomic on supported POSIX filesystems
• source hashes are checked again immediately before publication
• generated unittest wrappers verify the complete source manifest, including helpers, on every module resolution
• CRLF/LF differences are normalized only for the JSON manifest; source hashes remain byte-exact
• unsupported or ambiguous constructs fail closed with an actionable report
• this is filesystem transaction safety, not an OS security sandbox

Supported migration scope

Pytest support includes static module-level tests, one static parametrize, common static skip forms, plain markers, and a deliberately limited fixture subset. Unittest uses compatibility wrappers around real TestCase.run so lifecycle, assertions, mock, async cases, and per-test outcomes are preserved. Dynamic collection, plugins/hooks, class-based pytest suites, run-global session fixtures, and other constructs whose semantics cannot be guaranteed are blocked instead of guessed.

Reproducible benchmark snapshot

All rows use 3,000 tests across 64 modules, one warm-up, five alternating measured rounds, and four native workers:

Source runner / workload	Source median	Native median	Result
sequential pytest / no-op	1.539 s	0.521 s	Testenix 2.96x faster
sequential unittest outcome probe / no-op	0.161 s	1.192 s	Testenix 7.40x slower
sequential unittest outcome probe / 1 ms work	4.066 s	2.577 s	Testenix 1.58x faster

Migration is a one-time validation transaction and took 5.940–17.251 s in these synthetic cases. The raw JSON records include clean source commit 3a51a90, exact commands, versions, environment, every sample, integrity result, and publication status. These results are not presented as a universal advantage over pytest-xdist or real repositories.

Verification

• 201 passed (full suite)
• 115 passed (migration-focused suite)
• Ruff check and format
• mypy strict
• Pyright
• docs asset regeneration and freshness check
• Sphinx dirhtml build with warnings as errors
• source distribution and wheel build
• strict Twine metadata check
• wheel-contents check
• clean wheel and sdist smoke migrations for pytest and unittest
• CI matrix: CPython 3.11–3.14 on Ubuntu, macOS, and Windows

No PyPI release is performed by this PR.