docs: correct PR 592 evidence wording

[heyong4725]

Summary

Fixes the four findings from the Codex cross-review of #592:

• records the unattested M3 v2 measurement in the post-2026-08-28 delta
• labels technical-report section 9 as a historical 2026-08-16 snapshot and points later measurements to the canonical README
• narrows the lab briefing wrong-object statement to retained development summaries, matching the weakened catalog row
• documents explicit traceability waivers in both the lab briefing and glossary

Requirement IDs

CLM-5, CLM-8.

Gates

• ruff format --check .: pass
• ruff check .: pass
• claim-evidence checker: pass
• docs inventory checker: pass
• focused documentation policing tests: 97 passed
• pytest -m unit: local run reached 769 passed before one unrelated macOS dynamic-worker test returned infrastructure_exclusion; isolated diagnosis showed its child was killed when only 54.889 seconds remained of the 120-second tool budget after local runtime attestation. The identical test and complete unit suite pass on docs: refresh project status for 2026-09-14 and add a lab briefing #592 hosted macOS CI. Child-PR CI is the clean-environment gate.

No production code, tests, generated artifacts, specs, or frozen-set files changed.