docs: document video-digest + verify, sync stale video-digest render

[VGonPa]

Docs-only PR syncing the manual/reference docs to the enrichment features shipped on develop. No source or test changes; uv run --extra dev poe check stays green.

What shipped that the docs missed

• video-digest (PR feat(video-digest): xbrain video-digest step (long-form digest per video) #78, issue Long-form media digest: transcribe + summarize bookmarked videos (and podcasts) into the vault #44) — generates the long-form readable digest per video from its transcript + frame descriptions; worksheet flow (--executor manual|claude-code, --apply), writes data/video-digest-worksheet.json, auto-snapshots on apply.
• verify (PR feat(verify): xbrain verify — LLM-as-judge machinery for enrichment (PR-1 of #79) #80, issue Enrichment verification layer (LLM-as-judge evals for summary/digest/topics) #79) — LLM-as-judge over enrichment outputs (--target summary|digest|topics|all), --apply accepts multiple worksheets (one per judge), writes data/verify-report.{json,md}, report-only (never mutates the store).
• Video digest render (PR feat(generate): render long-form video digest + collapsible raw evidence #77) — the digest is now the section headline; raw transcript + frames are demoted into a collapsible <details> ("Frames + transcript" / "Frames y transcripción"); empty digest falls back to the old inline layout.
• enrich video inputs (PR feat(enrich): video frame descriptions + full transcript into enrich #75) — the worksheet track now sends the FULL transcript + a video_frame_descriptions field; only the api track truncates to TRANSCRIPT_CHAR_LIMIT.
• The additive digest: str = "" field on ContentSourceSuccess.

Changes

P0 — new commands documented

• README.md §Commands: video-digest + verify rows.
• README.md §Execution modes: added both to the LLM-stage list; noted they reuse [enrich].executor, worksheet tracks only (no api).
• docs/tutorial.md §5 + docs/digest-video.md §Run it: inserted the video-digest step before generate (+ optional verify).
• CLAUDE.md + ARCHITECTURE.md: video-digest and verify architecture bullets / stage sections; video-digest added to the items.json writers list + a verify-report.{json,md} artifact row.

P0 — corrected the stale ## Video digest render docs (digest headline + <details> demotion + empty-digest fallback): docs/digest-video.md §What you get, README.md §Layer 1, ARCHITECTURE.md §Video digest section.

P1

• Documented the additive digest field (ARCHITECTURE.md §x_video, CLAUDE.md).
• Fixed the "both enrich tracks truncate" claim → full transcript + frame descriptions on the worksheet track (ARCHITECTURE.md §enrich, CLAUDE.md); noted frame descriptions feed enrich/topics (README.md §Layer 1).

P2

• config.toml.example: [enrich].executor comment notes video-digest/verify also read it ([frames] untouched).
• docs/troubleshooting.md: video-digest/verify "no pending / worksheet" FAQ entry.

Every claim was verified against the code on develop (command flags, field names, file paths, i18n strings, snapshot labels).

🤖 Generated with Claude Code