Add safe pytest migration essentials for v0.2.0

[polishdataengineer]

Summary

• prepare Testenix 0.2.0 with native tmp_path, transactional monkeypatch, and autouse fixtures
• extend safe pytest migration to simple test classes, bare pytest.mark.asyncio, and statically provable local monkeypatch helpers
• preserve pytest-asyncio's default function-scoped loop lifecycle with an isolated asyncio.Runner per test or case
• fail closed on unsupported monkeypatch flows, class lifecycle, async configuration, custom loop policies, and ambiguous pytest config
• make migration summaries honest and compact while retaining complete JSON diagnostics
• update README, GitHub Pages content, changelog, CLI reference, roadmap, and LLM-ready documentation

Why

A real 118-test pytest suite was initially blocked by 161 unsupported constructs, mostly async markers, autouse fixtures, simple classes, tmp_path, and monkeypatch. Supporting those constructs safely requires native runtime equivalents and semantic guards; simply stripping pytest markers would change event-loop and fixture behavior.

User impact

Existing pytest sources remain untouched. testenix migrate pytest ... --check now converts the supported subset in private staging, compares source, native serial, and native parallel outcomes, and leaves the destination absent unless publication is explicitly requested and every gate passes.

Validation

• 284 passed
• Ruff check and format: clean
• mypy and pyright: clean
• generated docs check and Sphinx -W: clean
• lockfile check: clean
• wheel and sdist built as testenix 0.2.0
• Twine strict and wheel-content checks: clean
• isolated wheel/sdist smoke tests and migration examples: passed
• real project: 118/118 passed under pytest, native serial, and native 4-worker validation; originals unchanged and check output absent
• independent multi-agent re-audit: no release blockers

The observed real-project validation timings were 3.120 s (pytest), 2.870 s (native serial), and 2.423 s (native parallel). These are validation timings, not a multi-round benchmark claim.