feat(experiments): validate reproducible performance records

[ZacharyZcR]

Summary

• add a dependency-free validator for reproducible experiment manifests
• require a single changed variable, at least three raw samples, a correct median, and an explicit quality gate
• bind raw evidence with SHA-256 and retain negative or no-change results in the same format
• document the workflow in CONTRIBUTING.md and the pull-request checklist

Why

Performance claims are difficult to review when the changed variable, raw samples, aggregation, quality result, or source revision is missing. The manifest makes those fields machine-checkable without adding a runtime dependency.

Validation

• rebased onto current dev
• PYTHONPATH=. python3 tests/test_experiment_manifest.py: 5/5 passed
• python3 experiment_manifest.py ../docs/experiments/manifest.example.json: passed
• git diff --check: passed