docs+ui: 'verify this review' — public reproducibility walkthrough + measured-accuracy display

[JSONbored]

Problem

Epic #8211 track G capstone. The reproducible-backtest story (#8136's conclusion) and the public precision numbers only build trust if a skeptic can actually FOLLOW them: a docs page that walks re-running a published corpus snapshot to the same checksum, and a UI surface rendering the measured-accuracy numbers.

Requirements

⚠️ Required pattern. Fumadocs entry per the standing checklist (mdx + navs + route-count pins where applicable); UI components in the main site's design system (ui-kit tokens — text-token-*, border-hairline; raw Tailwind sizes are lint errors); data ONLY from the public precision endpoint — never a second computation path.

• Docs: end-to-end walkthrough — download snapshot, verify checksum, re-run scoreBacktest/compareBacktestScores, compare — with the exact commands, plus an honest 'what this does and does not prove' section (it proves reported scores are real; it does not prove the live gate ran this code — link the TEE disposition for that boundary).
• UI: an accuracy section on the public stats surface rendering per-rule precision with the sample-floor null state designed deliberately ("insufficient data" never renders as 0%).

Deliverables

• Docs page wired into navs + drift-guard pins updated
• UI section with loading/empty/null states + screenshots per the visual-PR convention

Links & Resources

#8211 (epic), the precision-endpoint sub-issue (blocked-by), apps/loopover-ui/content/docs/backtest-calibration.mdx, #8136/#8137 (the TEE boundary to cite honestly)

Boundaries

Render the endpoint; compute nothing client-side beyond formatting. No corpus content.

[linear-code]

JSO-5715