Tier 1S of #150 — targets SPA route transitions specifically. Without these, a React/Vue route swap that mounts thousands of nodes either waits out a 250ms throttle window before the new content is hidden, or misses URL-gated selectors that newly apply to the route.

Summary

lib/route-change.ts (new) — process-wide route-change emitter:

• Hooks the Navigation API navigatesuccess (modern Chrome) plus popstate and hashchange as belt-and-suspenders.
• Dedupes by URL so multiple events for the same navigation fire the listener once.
• Older browsers without Navigation API miss pushState-only routes — but the watcher's normal MutationObserver still catches up. This signal is a latency optimization, not a correctness requirement.

lib/subtree-watcher.ts — wire route changes + detached fast-path:

• Subscribe to subscribeRouteChange in start(). On a detected change:
  1. Cancel the pending throttle, discard buffered MutationRecords, clear pending.
  2. Schedule a one-frame-later sweep that calls onSubtrees([document.body]).
  3. The rAF wait lets the framework commit the new route's DOM before we scan.
  4. Consecutive route changes coalesce — only the latest scheduled sweep fires.
  5. Skipped while the tab is hidden or the watcher has been stopped.
• Detached-subtree fast-path in enqueue(): React reconciliation routinely adds-then-removes nodes within one synchronous block. By the time the MO callback fires, the addition record is still there but the node is already detached. drain() filters by isConnected too — checking at enqueue keeps pending small during noisy route-swap bursts.

What Tier 1S items are landed

From issue #150:

• ✅ ci: stop double-zipping extension artifact #7 Route-change signal as proactive coalesce
• ✅ docs: scaffold Astro Starlight documentation site #8 Pause-and-batch around route transitions (via the rAF deferral)
• ✅ build(deps): Bump actions/upload-pages-artifact from 3 to 5 #9 Detached-subtree fast-path

Test plan

• bun run test → 1177 passed (1164 baseline + 6 route-change.test.ts + 7 new subtree-watcher cases)
• bun run typecheck
• bun run check (biome + eslint)
• bun run knip
• bun run test:coverage — statements 85.55% / branches 76.81% / functions 88.71% / lines 85.46% (above ratchet)
• Manual smoke on a real SPA (e.g. a React Router app, GitHub web UI) — confirm placeholders apply on initial route AND after a route swap, no double-application, no perf regression on rapid back/forward

Refs #150.

🤖 Generated with Claude Code