Sprite Lab in Lab2: fix level switch

[breville]

[written by Claude]

Navigating between two SpriteLab2 levels in a lesson crashed the page: Error: Can't change value of key "channelId", then the error boundary.

Root cause

SpriteLab2 seeds the legacy pageConstants slice (isBlockly, channelId, …) because it bypasses StudioApp.init, and the legacy reducer forbids changing a key once set — it assumes one page load per level, and offers no reset action. Lab2 switches levels in place: the second level seeds a different channelId and the reducer throws mid-dispatch. Single-level sessions never hit it because the first undefined → value transition is legal; it takes a second, different channel.

The fix

The lab registers the reducer wrapped (reseedablePageConstants): the wrapper handles a lab-owned RESET_PAGE_CONSTANTS action (returning the legacy initial state) and delegates everything else untouched, guard intact. The seeding effect dispatches the reset from its cleanup, so React's cleanup-before-effect guarantee means every seed — including the next level's after an in-place switch — lands on a fresh slice. The legacy "constants never change" invariant stays enforced between seed cycles; a test proves a changed channel without a reset still throws.

This shape is only safe now that useSources (#73868, merged into this branch) owns the channel: the inner view unmounts while the next level's project loads and the channel arrives as a fresh prop on remount, so nothing can re-seed a stale channel between the reset and the seed.

Why not reset on LifecycleEvent.LevelChangeRequested

Tried first — it's the formal seam, and the event does fire. But the seed effect re-runs when progress.currentLevelId changes, while the old view is still briefly mounted with the old channel: it re-seeds the old value right over the event-ordered reset, and the throw comes back when the new level seeds. Tying the reset to the seed effect's own cleanup is immune to this: reset and seed are paired by React's effect ordering rather than racing through event timing.

(An earlier revision of this PR used a catch-and-retry wrapper — delegate to the legacy reducer, re-apply from initial state when its change-guard throws — because before #73868 the stale-channel race made any ordered reset unwinnable. With the race gone, the explicit reset replaces the exception-as-control-flow.)

Testing

• Unit tests on the wrapper: reset/seed cycle accepts a new channel; a changed channel without reset still throws; disallowed keys still throw; unrelated actions delegate.
• Verified in-browser on two adjacent SpriteLab2 levels: switching 9 → 10 → 9 in-place shows a distinct channel per level, the original channel restored on return, workspaces rendering, no error boundary, no reducer errors.