test(workbench): add React Testing Library coverage for autosave store-swap timing

[chrisjmendez]

Intent

Three autosave bugs were introduced and fixed in PR #66 — all living in a gap that types, lint, and the existing unit tests cannot see: React effects in the same render don't observe state set by sibling effects in that render. A single integration test with @testing-library/react would have caught every one of them before any data was lost.

The three bugs this test would have caught

#	What went wrong	Observable symptom
1	useProjectPersistence load effect had [projectId] as its only dep. When Clerk resolved and swapped localStore → apiStore, the load never re-ran. storeRef.current intentionally defeats react-hooks/exhaustive-deps.	Existing project always loaded empty canvas
2	hydrated.current never flipped true when initialState === null (new project). Logic error; types were correct.	Autosave never fired; state lost on every reload
3	Save effect fired with stale canvasNodes = [] in the same render where setCanvasNodes([...real nodes...]) was called by the hydration effect. State from setCanvasNodes isn't visible until the next render. exhaustive-deps only catches missing deps, not timing-wrong extras.	0-node PUT overwrote a real row on hot-reload unmount

What's needed

1. Add @testing-library/react (+ @testing-library/user-event) to apps/web devDependencies.
2. Mock useUser from @clerk/nextjs and IProjectStore so tests control Clerk's load timing and store responses.
3. Write the core assertion test:

   it("save never fires with fewer nodes than initialState after store swap", async () => {
     // mount with localStore → returns null
     // simulate Clerk resolving → swap to apiStore with 6-node project
     // assert: store.save never called with canvasNodes.length < 6
     // assert: store.save eventually called with canvasNodes.length === 6
   })

4. Add two supporting cases: (a) new project — save fires only after user adds first node, not on load; (b) reload round-trip — mount → load N nodes → unmount → remount → same N nodes loaded, no save in between.
5. Wire into pnpm test (already runs vitest; RTL tests belong in tests/seed-validation/ or a new tests/hooks/ directory).

Acceptance

• pnpm test is green and includes at least the three cases above.
• No eslint-disable react-hooks/exhaustive-deps comment is needed after the refactor (the hooks should be lint-clean by design).
• A future regression that re-introduces any of the three bugs causes at least one of these tests to fail red before merge.

Links

• PR #66 — spec 028 workbench autosave — the PR where all three bugs were introduced and fixed; commit history has the exact diffs
• apps/web/src/lib/workbench/hooks/use-project-persistence.ts — hook under test; load effect deps are the core invariant
• apps/web/src/app/workbench/workbench-canvas.tsx — hydration + save effects; the timing interaction between them is what needs covering
• tests/seed-validation/use-project-persistence.test.ts — existing unit tests (store + debounce logic only; no React rendering)
• specs/028-workbench-project-autosave.md — spec defining the persistence contract these tests should verify

---

Originally tracked internally as Haptic-AI/festivus#67.