[codex] Align public copy with evidence registry semantics

[toby-bridges]

Summary

• Replace remaining public leaderboard / relay-use recommendation wording with evidence-registry and needs-review language.
• Update homepage demo copy, comparison docs, OpenClaw/Hermes skill summaries, and report text so LOW/clean-looking results are not framed as safety certification.
• Add a regression check that public growth surfaces do not drift back to leaderboard/safe-reliable wording.

Validation

• python3 scripts/build-standalone.py --check
• python3 scripts/collect-metrics.py --check
• python3 scripts/sync-version.py --check
• python3 -m pytest tests/test_skill_distribution.py tests/test_process_submission.py tests/test_example_report.py -q
• python3 -m pytest tests/test_dual_distribution_parity.py tests/test_static_site.py -q
• git diff --check
• python3 -m pytest -q (776 passed)

Notes

• No audit detector logic changed; the standalone artifact was regenerated because two user-visible report strings changed in scripts/audit.py.
• This keeps the contribution pipeline framed as a review-gated evidence registry, not a relay ranking or vendor endorsement.