docs(perf): retain native baselines and exercise documentation CI

[jewei]

Documentation/performance follow-up

Publish sanitized, source-identified native baseline measurements and package hashes from verified #10, and link the measurement procedure to that record. Distinguish settled DOM, double-rAF rendering opportunity, and actual paint; no application behavior changes.

Independent reviews

Two fresh independent reviewers, a10e1c03-99ae-4d5 and 0f6ced7d-b0d6-43f, approved the numeric/provenance review. Each recalculated all 16 published p95 values from retained raw samples and checked environment, source/tree identity, and package hashes. These are subagent reviews, not human GitHub approvals.

After integrating protected main, the documentation patch remains identical to reviewed f17507a (stable patch ID 89e9dabe84ab95570b62849b65a84d91189f4b21). The historical version-1 table is not a claim about latest-version performance; the current procedure documents the version-2 readiness boundary and warns against cross-procedure comparisons.

Verification

Repository/documentation and formatting checks passed. Earlier hosted run 36379520980 proved the docs-only gate: repository/change detection/Required checks passed, and quick/desktop/native jobs were intentionally skipped.

Final current-base run 36390647539 passed repository checks, change detection, and the protected Required checks gate for 18d05a3. Quick, desktop, and native jobs were intentionally skipped for the two documentation-only files. Local bun run check:repo (250 files, 41 documents) and bun run format:check also passed after the final main integration.

No new desktop, physical-paint, memory, or latency-target proof is claimed by this docs-only change.