[P3.22] Dashboard renders a 100-book page with no pagination UI; first paint exceeds 15s on a large library

[frankbria]

Surfaced while triaging #488.

What happens

GET /api/v1/books/ returns up to 100 books (limit: int = Query(100, ge=1, le=100)), and the dashboard renders every one as a card. There is no pagination control in the UI, so a user past 100 books simply cannot reach the rest.

Measured on staging, where the E2E account holds 2,702 books:

• 100-book payload: ~195 KB (~1.9 KB per book)
• 100 cards rendered in one list
• First paint exceeded 15 s late in a suite run — an E2E spec waiting 15 s for its book to appear captured a snapshot with zero cards rendered

That spec has been moved to 20 s to match its siblings (#488), which treats the symptom. The underlying slowness is this issue.

Why it is worth fixing

• The remaining 2,602 books are unreachable. skip exists on the endpoint but nothing in the UI drives it.
• A ~200 KB payload plus 100 card renders is a poor first-paint budget for the app's primary landing screen.
• Ordering only recently became correct (fix(books): show newest books first on the dashboard #491 added sort(updated_at, -1); before that the page showed the oldest 100 forever). With ordering fixed, the missing pagination is now the binding limit.

Notes for whoever picks this up

• owner_updated_idx (owner_id, updated_at DESC) already exists and backs the current sort — explain() shows LIMIT → FETCH → IXSCAN. Keep any paging change on that index; adding an _id tiebreaker measured as a blocking in-memory SORT stage.
• Cursor-based paging (updated_at + _id as a cursor) avoids the skip/limit drift problem on a list that changes while you page through it, but note the index caveat above.
• The list endpoint already projects out per-chapter content, so the payload is metadata only — the size is card count, not fat rows.

Priority

P3.22 — degraded UX and an unreachable tail, not a correctness or security problem, and it only bites accounts with large libraries. The staging E2E account is the extreme case; real users are unlikely to be near 100 yet.