fix(pokedex): open detail cards without a per-card round trip

Since the grid transport change, opening a Pokémon fetched
/api/pokedexes/[id]/entries/[entryId]. That request re-scanned the whole
dex to authorise one entry, on top of an auth call and an ownership
query, and production runs its functions in a different region from the
database — so each card took seconds to fill in. The page's 60-second
reconcile also wiped the detail cache, making cards refetch about once a
minute.

Details are catalog text plus personal notes, so the dex is now read once
per account/dex in the background at the interactive/idle boundary and
fills that cache; a card opens straight from memory. That read reuses the
existing combined-data endpoint, which gains an opt-out for its count
query since a caller taking every row already knows the total.

The per-entry endpoint stays as the fallback for a card opened before the
background read lands, and now checks membership for the single entry
instead of materialising the dex. Catch status still comes from the live
grid row and any queued write, so a cached detail cannot show stale
progress.

The packed grid transport and virtualised rendering are unchanged.

docs/performance/pokedex-implementation.md

@@ -7,7 +7,8 @@ Status: application changes implemented; hosting comparison remains gated. Measu
 - Ownership and saved scope links are read together. Scoped entry retrieval is shared by rows/count consumers; the full grid performs no count query. Catch joins use ID maps. Scope deduplication, named default forms, supplements and ordering remain covered by repository tests.
 - The page awaits the entire compact grid and renders initial boxes on the server. Authentication state reaches SSR through the validated layout user. Successful navigation needs no grid API request; failed grid loads expose retry.
 - `PokedexGridRow` contains identity, sprite resolution fields and catch flags. The page and authenticated `/api/pokedexes/[id]/grid` endpoint transport named tuples defined in `PokedexGridRow.ts`; `packGrid`/`unpackGrid` keep that wire format out of components. Instructions, notes, origin games and repeated owner/dex IDs are absent. Existing full combined-data consumers retain their contracts.
-- The detail endpoint verifies ownership and membership before returning one full `CombinedData` row. The modal opens immediately with identity and full artwork, then loads editable details. Its account/dex/entry cache, abort/sequence checks and pending-patch merge protect rapid selection changes and optimistic edits.
+- Details for the whole dex are read once per account/dex in the background, at the same interactive/idle boundary as the other deferred startup work, and fill the modal's cache. Opening a card therefore renders from memory with no request of its own: a per-card round trip cost seconds on the deployed cross-region setup, which local timings did not show. That read reuses `/api/pokedexes/[id]/combined-data`, which now skips its count query when a caller asks for every row at once.
+- The per-entry detail endpoint remains as the fallback for a card opened before the background read lands, and for a failed read. It verifies ownership and membership before returning one full `CombinedData` row, checking membership for the single entry rather than materialising the dex. The modal opens immediately with identity and full artwork, then fills in editable details. Its account/dex/entry cache, abort/sequence checks and pending-patch merge protect rapid selection changes and optimistic edits; catch status always comes from the live grid row, so a cached detail can never show stale progress.
 - Status writes send changed fields. Bulk writes group records by supplied columns, preserving omitted notes and flags; explicit empty notes clear them. New records use database defaults. Full-record callers and exports remain supported. Bulk box actions still target all original 30 slots, including dimmed entries.
 - Grid placeholders preserve geometry; visible boxes and one row of overscan mount populated cells. Focused boxes remain mounted, keyboard navigation crosses boundaries, modal close restores focus, and an accessible render-all option exposes the complete document. Each Pokémon uses one button with identity/status and a noninteractive tooltip.
 - Density is persisted in a cookie for stable SSR geometry and in local storage. Resize/density changes preserve the current box anchor. Existing local-storage-only preferences are replaced by the cookie after choosing a density.
@@ -26,7 +27,7 @@ Fixtures: national (1,025 entries), Scarlet/Paldea forms (439 entries), mixed ca
 
 The packed grid is approximately 81 KB national and 36 KB scoped, versus the investigation's approximately 593 KB and 255 KB full-data JSON. The same-fixture integration assertion separately verifies at least 60% reduction against full combined rows. These are serialized row sizes, not compressed HTML document sizes.
 
-The local browser checks verify at most 180 populated mounted cells, fewer than 2,500 DOM elements and CLS at most 0.1; observed initial population is 120 cells. Density/mobile checks include all three densities at 1,350 and 390 pixels. Detail artwork is asserted to load at 512 pixels. No redundant grid request is allowed; intentional detail requests are allowed.
+The local browser checks verify at most 180 populated mounted cells, fewer than 2,500 DOM elements and CLS at most 0.1; observed initial population is 120 cells. Density/mobile checks include all three densities at 1,350 and 390 pixels. Detail artwork is asserted to load at 512 pixels. No redundant grid request is allowed; the dex is read for details exactly once, and opening a card afterwards must make no request at all. The offline-isolation checks block that background read so the snapshot path, not the primed cache, answers those clicks.
 
 Warm results from the corrected harness (30 samples per row; [sanitized summary](pokedex-local-results.json)):