perf(reader): coalesce relocate events and memoize BookCell to stop per-swipe storm

[loveheaven]

Summary

Two complementary changes that stop a swipe burst from re-rendering the full reader tree on every intermediate relocate.

FoliateViewer — coalesce relocate

foliate fires relocate multiple times during a swipe burst (snap steps plus the final stabilize). Each one ended up in setProgress, which writes to readerProgressStore plus bookDataStore. This change coalesces them into a single commit per animation frame via requestAnimationFrame, so only the final viewport state is persisted.

An earlier iteration used requestIdleCallback, but profiling on Android showed "Fire Idle Callback" ballooning to ~2 s of total time per ~28 s session: rIC backed up under sustained pressure and dumped the whole queue into the post-swipe pause, producing exactly the "feels sluggish right after I let go" jank the change was trying to fix. rAF runs once per frame, gets scheduled by the browser's normal vsync loop, and doesn't accumulate when the page is busy.

BooksGrid → BookCell

Previously BooksGrid subscribed to the entire progresses map and rendered all per-book chrome inline, so every page turn re-rendered every cell in the grid (cheap but multiplicative when parallel views are open).

Extract a memoized BookCell component that:

• subscribes only to its own book's progress via useBookProgress(bookKey) — a page turn re-renders one cell;
• consumes per-field store selectors so action identities are stable and React.memo doesn't get defeated;
• receives memoized gridInsets / contentInsets from BooksGrid so memoization on prop identity is meaningful.

Why this matters

Pairs with the progress-store split (#4557). #4557 isolates the high-frequency write source from the wider reader store; this PR makes sure the now-tight reactive surface only fires on full-frame transitions and that only one cell at a time gets the update.

Depends on #4557

This PR's first commit duplicates #4557 (split progress store) so the second commit can compile in isolation (it uses useBookProgress). Once #4557 lands, GitHub will collapse the diff to only the coalesce / BookCell changes. Please merge #4557 first.

Testing

All existing tests pass (pnpm test — 5288 passed).

[chrox]

Reviewed and pushed a follow-up fix (714c10a0d).

Issue: the BookCell extraction subscribes reactively only to useBookProgress(bookKey) and reads viewState/viewSettings imperatively. On current main the grid re-rendered on any readerStore change (whole-store destructure subscription), which is what kept the per-book chrome in sync. After this PR, per-book state that doesn't flow through the progress store no longer re-renders the cell:

• Show Header / Show Footer and Double Border / Border Color save via saveViewSettings(..., applyStyles=false) → no setStyles → no relocate → no progress write.
• Bookmarking writes setBookmarkRibbonVisibility → viewStates[key] only, no progress.

Since SectionInfo / ProgressBar / Ribbon are mount-gated inside BookCell (and don't self-gate), toggling those left stale chrome until the next page turn.

Fix: BookCell now also subscribes to its own viewStates[bookKey] slice. This is safe precisely because #4557 moved progress into its own store — viewStates[key] now bumps only on low-frequency events (settings toggles, ribbon, init, sync), never on the per-swipe relocate path — so it doesn't reintroduce the commit storm the split removed.

The FoliateViewer rAF coalescing looks correct (stable refs/callback, register-once wiring, unmount flush; previewMode clearing is on the untouched renderer-relocate path).

pnpm test (5288 passed) and pnpm lint (tsgo + Biome) pass.

Minor, left as-is: ReaderContent's handleCloseBook / handleCloseBooksToLibrary aren't useCallback'd, so React.memo(BookCell) is defeated whenever ReaderContent re-renders. It's off the per-page-turn path, and memoizing it correctly also requires stabilizing the throttle(...), so I left it.