docs([issue-2073]): rewrite four shipped feature docs from plan-tense to verified current-state

[atomantic]

Summary

Closes #2073.

Four feature docs described shipped features but still read as pre-build plans/specs. Each is rewritten as current-state documentation with every claim verified against the code (via parallel research agents + a three-reviewer series loop):

• docs/features/digital-twin.md — "Planned Phases" reconciled into feature sections. Corrections: imports are Goodreads/Spotify/Letterboxd/iCal (no Last.fm); multi-modal capture is pasted-transcript/image only (no live voice/video); style comparison is qualitative; one active persona overlays the base twin; feedback weight adjustments are advisory-only (recorded, not yet consumed); all four test suites run per-persona; persona routes split accurately (no GET by id). An honest "Not Built" section records the unshipped roadmap ideas.
• docs/features/openclaw-operator-chat.md — proposal sections (Proposed capabilities / Open questions / Recommended design decisions / Initial deliverables) trimmed into a living doc covering the shipped 5-endpoint surface, file/env config (full key set), and graceful degradation. Keeps the honest proposed-but-unbuilt list (POST /sessions, /jobs, /subagents, /context/compose).
• docs/features/post.md — "Existing vs New (M55)" spec framing removed. Adds the undocumented Cognitive domain, Morse training (training-log only), drill cache + consent flow, training log, adaptive difficulty, spaced repetition. Corrections: drill-type ids (doubling-chain, pun-wordplay), memory drills aren't launched in scored sessions, full-vs-quick session composition, cognitive drills are untimed, session score is an unweighted mean, real component inventory.
• docs/features/writers-room.md — implementation plan (PR feat(creative-director): autonomous long-form video projects + media job queue #191 baseline, Phases 0–5) converted to current-state: PostgreSQL-primary storage via store.js (Migrate Writers Room metadata to PostgreSQL (.md bodies stay file-backed) #1017) with file-primary draft bodies, CD bridge (render-plans were never built), story bible, storyboard suite (incl. Adapt's auto-queued renders), live-director budgets, federation record kinds vs sync categories. The "Synchronized Review Surface" section that syncedReview.js cites is preserved and now describes the actual derive-on-read mapping model.

Also: soul-system.md / digital-twin.md / identity-system.md now cross-link with a note on which angle each covers (document workflow vs quantitative modeling vs identity architecture) — the issue's optional consideration, done as cross-linking rather than consolidation.

Test plan

• Full server suite green in the worktree: 15,659 passed / 154 skipped / 0 failed (the issue's explicit caution — planIds.test.js and the writersRoom services only reference the doc in comments/string literals; nothing depends on its content).
• Three-reviewer series loop (claude, ollama, codex) run to a clean converged verdict; every reviewer finding was verified against source before fixing.