feat(core/frontend): ai-tools per-mode layouts (image=chat-stream / model=person picker / cover=platform) + pipeline real unread bell

- ImageWorkbenchPage now renders mode-specific layouts matching each baseline:
  image -> chat-stream (conversation list + hero + prompt chips + chat input bar);
  model -> product rail + 真人模特 cards (assets category=person, fallback Ava/Luna/Mia/Zoe) + per-model count;
  cover -> platform-kit picker. Generation (onGenerate) wiring + loading/empty/fail states preserved.
- pipeline.tsx: bespoke topbar bell now shows real unreadCount (was hardcoded 12); App.tsx threads it.
verified: tsc --noEmit clean; screenshot confirms image-optimize matches chat-stream baseline.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

core/frontend/src/App.tsx

@@ -463,6 +463,7 @@ export function App() {
         assets={assets}
         billing={billing}
         notice={notice}
+        unreadCount={unreadCount}
         avatarChar={avatarChar}
         logout={logout}
         onRefresh={refreshProjectDetail}