feat(web): scripted full-screen product demo mode

[vibegui]

Summary

A public, fully-mocked, full-screen product demo (oya.ai-style) that animates the real Studio chat components from a scripted stream — no backend, MCP transport, or auth. The whole thing is a reusable framework: a demo reads top-to-bottom like a storyboard.

await d.user("Here's my store: https://acme.com — make it faster");
await d.tool(takeScreenshot({ url, image }));
await d.stream(DIAGNOSIS, { instant: true });

Two demos, each on its own URL (light mode)

• /demo/storefront — business user drops in a URL → accurate perf+SEO diagnosis scorecard (real GFM table) → fix plan → parallel fixes → re-audit before/after → deploy.
• /demo/agents — Studio as a web Conductor: the ghost cursor clicks Connect desktop → the real link modal opens → a mocked iTerm window runs `bunx decocms link` then minimizes → modal flips to Connected → work runs across two orgs in parallel (chat + live preview), switching context while the first org keeps building in the background.
• /demo — a chooser linking to both. No cross-scenario auto-advance.

How it works (apps/mesh/src/web/demo/)

• Director — non-React class owning all timing via `@decocms/std` `sleep`; mutates external `Store`s that components read through `useSyncExternalStore`, so no `useEffect` is needed to animate (the one lifecycle effect is documented, same pattern as `live-timer.tsx`).
• Chat seam — `DemoChatStreamProvider` supplies a scripted `ChatStreamContextValue` (via a new exported `DemoChatStreamContext`), so the real `Chat.Messages` → `MessageAssistant` → tool-call renderers paint it identically to a live stream.
• DemoProviders — mock `ProjectContext` + a network-free `QueryClient` (no auth gate).
• Multi-track chats, org switching, iframe previews, ghost cursor, typed terminal — all driven by the same Director primitives.

The autoplay runner is guarded by a per-`stores` singleton with deferred teardown, so the dev runtime's spurious passive-effect re-runs (React Compiler / StrictMode) can't abort + restart a scenario mid-play.

Affected areas

• New: `apps/mesh/src/web/demo/**`, `routes/demo.tsx`, `routes/demo-scenario.tsx`
• Touched: `web/index.tsx` (register `/demo` + `/demo/$scenario` as public routes), `components/chat/chat-context.tsx` (export the chat-stream context for the demo seam — one line)
• No backend/server changes.

Testing

• `bun run check` ✅ · `bun run lint` ✅ · `bun run fmt:check` ✅
• Verified end-to-end with Playwright: both demos run to completion in a stable loop with 0 console errors; zero network requests beyond static assets.
• Screenshots: view live at `/demo/storefront` and `/demo/agents` (dev). UI is fully scripted/animated; static frames don't capture the motion.

🤖 Generated with Claude Code

---

Summary by cubic

Adds a public, full-screen Demo Mode that scripts the real Studio chat UI with no backend or auth. Now includes inline sprint/PR/daily-digest cards and an agent sidebar with background notifications; each demo plays once and then shows a CTA end card.

• New Features

  • Inline chat cards: work plan (approve then live task ticks), pull request (checks → merge), and daily digest (connect Slack/Teams) via a tiny part-renderer registry.
  • Storefront demo: capture page → audit → show diagnosis → approve sprint → PR checks+merge → deploy → re‑audit → connect daily digest.
  • Agents demo: Studio-style sidebar to switch agents; work continues in the background with a notification dot; connect-desktop dialog + mocked iTerm and live preview panes.

• Refactors

  • Core chat accepts external part renderers before its built-in switch; demo registers its cards, app behavior unchanged.
  • Exported DemoChatStreamContext; updates to card outputs are applied immutably so memoized components re-render reliably.

^{Written for commit 813d948. Summary will update on new commits.}