Harden issue 122 containment checks

[stef-k]

Summary

• restore fresh physical containment validation for baseline-enabled analysis
• replace the benchmark harness with a testable Python implementation that rejects lexical and physical output containment before writes
• fail closed when pre-run or post-run Git status verification fails
• align usage documentation and changelog wording

Validation

• independent read-only review: READY FOR FIXING PR at 759456dd8fc641592fcb2670013bc590dd29da55
• focused remediation suite: 58 tests passed (32 containment/baseline tests plus 26 production-analysis tests in fresh orchestration validation)
• implementation fresh installed-wheel discovery: 326 tests passed
• fresh wheel/sdist build and isolated installation: passed
• repository dogfood: expected REVIEW under --ci
• Python harness syntax/help smoke and git diff --check: passed
• retained Wayfarer performance evidence unchanged; no expensive benchmark rerun

Follow-up to #122 and #123.

[stef-k]

Final bounded harness smoke at exact head 759456dd8fc641592fcb2670013bc590dd29da55:

• exercised the Python harness end to end against the pinned disposable Wayfarer checkout using an external smoke-stub analyzer (no production benchmark rerun)
• all LOC-only, syntax-only, and normal warmups plus three samples completed successfully
• cProfile subprocess and consolidated benchmark-results.json completed successfully
• pre/post Git status matched; normalAnalysisCreatedRepositoryMetadata was false
• smoke artifacts retained outside both repositories under the OS temporary directory

Exact-head CI is fully green.