feat(maic-editor): enablement infrastructure (pre-slide-surface)

[wyuc]

Implements #570. Lands before #562. Stacked on #568 (the diff against
feat/maic-editor-v0 includes #568's commits until that merges).

What ships

1. Feature flag — NEXT_PUBLIC_MAIC_EDITOR_ENABLED, default OFF.
   Pro toggle in Header only renders when ON. StageMode unchanged.

2. Slide DSL schema versioning — SlideContent.schemaVersion?: number
   (current = 1). migrateSlideContent / migrateScene are pure +
   idempotent; the stage store funnels setScenes / addScene through
   migrateScene so legacy data gets stamped at the boundary.

3. Round-trip test harness — tests/edit/round-trip/. Caveat: no
   PPTX → Slide reimport exists in the codebase, so the harness does
   apply ops → buildPptxBlob → JSZip parse → assert content survived
   instead of a full reimport diff. Per-op cases extend it in [MAIC Editor] Slide surface — first concrete static-display surface #562.
   buildPptxBlob is now exported (the hook is still the only runtime
   caller).

4. Edit history persistence — per-scene persist / load / has /
   clear helpers under maic-editor:slide-history:${sceneId}, plus a
   standalone SlideHistoryRestorePrompt dialog. 4 new i18n strings × 6
   locales. Not wired into Stage yet — [MAIC Editor] Slide surface — first concrete static-display surface #562 owns the entry effect.

5. Concurrency guards —

   • isSceneEditLocked predicate for the AI-regen lock (defensive;
     current call paths can't structurally trigger it).
   • tryAcquireEditLock / refreshEditLock / releaseEditLock /
     isEditLockHeldByOther for multi-tab, localStorage-backed; atomic
     acquire, idempotent heartbeat, stale-lock takeover after 3×
     heartbeat (default 15s).
   • Standalone MultiTabEditConflictPrompt. 3 new i18n strings × 6
     locales. Not wired into Stage yet — same reason as (4).

Why dialogs aren't wired here

The slide-surface PR owns the edit-entry effect machinery — that's where
the history-state lifecycle and the per-tab tabId ref naturally live.
Shipping half-wired dialogs now means speculatively building Stage state
we'd restructure on contact with the surface.

Out of scope

Server-side history sync (Phase 2); cross-device conflict resolution;
optimistic regen guard.

Test plan

• prettier · lint (0 new warnings) · tsc · check:i18n-keys · build — green
• pnpm test — 410 passing / 9 failing; the 9 are baseline
  tests/server/ssrf-guard.test.ts (DNS-dependent). +45 new tests
  across feature-flag (5), slide-schema (9), round-trip (2),
  slide-history-persistence (8), regen-lock (4), edit-mode-lock (17).
• Manual: flag unset → no Pro toggle; flag set → Pro toggle behaves
  as in refactor(maic-editor): drop EditModeSidebar; clean Pro mode chrome #568. No new user-visible affordances.

🤖 Generated with Claude Code

[wyuc]

Self-CR pass

Ran a code-review loop. Verdict: READY. Findings:

• Feature flag correctness — isMaicEditorEnabled parses
  'true'/'1' only; everything else (unset, 'false', 'yes')
  falsy; tests cover all branches. Stage call site at
  components/stage.tsx:1025 passes undefined when off; Header
  gates the Switch on onToggleEditMode &&. ✓
• Schema migration purity + idempotency — same-reference return
  on already-current input; new object preserving canvas otherwise;
  non-slide scenes pass through identity. Store funneling in
  setScenes / addScene short-circuits via identity equality. ✓
• Forward-compat fix (CR follow-up commit `0c1b02c`): content
  written with a future schemaVersion >= CURRENT is now returned
  untouched instead of silently downgraded.
• Round-trip harness honesty — noop test unzips and asserts
  ppt/slides/slide1.xml present (not just blob size); needle test
  uses a distinctive string and asserts .toContain(NEEDLE) against
  the actual slide XML. ✓
• Storage failure handling — all four persistence helpers wrap
  storage access in try/catch; tests cover corrupted JSON and
  quota-exceeded. Multi-tab lock helpers same. ✓
• Multi-tab lock semantics — self-renewal allowed; refuses fresh
  other-tab lock; steals after LOCK_STALE_MS; release no-ops
  when another tab now owns; refresh no-ops on other-tab lock.
  Tests cover all branches. ✓
• No new upstream coupling — new lib/edit/ modules touch only
  the public types (Scene, SlideContent, StageMode,
  SlideEditHistory); EditShell imports unchanged.
• Not-wired claim is honest — both SlideHistoryRestorePrompt
  and MultiTabEditConflictPrompt are unmounted in this diff. [MAIC Editor] Slide surface — first concrete static-display surface #562
  owns the entry effect.
• i18n alignment — 7 new keys × 6 locales; check:i18n-keys
  passes.

Tests: 46 new (feature-flag 5 + slide-schema 10 + round-trip 2

• slide-history-persistence 8 + regen-lock 4 + edit-mode-lock 17),
  all green. 9 pre-existing `tests/server/ssrf-guard.test.ts` failures
  are baseline (DNS-dependent, sandbox-only).

Ready for review.

[cosarah]

LGTM. Locally verified:

• Feature flag gates Pro toggle (with/without NEXT_PUBLIC_MAIC_EDITOR_ENABLED)
• 45 new tests pass (lock, schema migration, history persistence, regen lock, round-trip)
• Confirmed lock/history APIs are infra-only with no runtime callers yet — matches PR scope note (#562 owns Stage wiring)

Code quality: lock atomicity and stale-takeover logic are sound; migrateSlideContent is idempotent and forward-compatible; localStorage paths are SSR-safe and exception-tolerant. No blocking issues.