v2.2.8.2

What's Changed

• Ships the large-result-set query overhaul (paging/streaming), search-ranking fixes, secure thumbnails, and the VFB-maintained griddle fork to production. by @Robbie1977 in #1705

Highlights

Large query result sets no longer capped

Query results now stream in pages instead of stopping at the old 10,000 /
50,000 ceilings, so full sets load (e.g. List all available images for a
broad class returns its true count — ~500k rows — rather than truncating).

The client pages by loaded-count and only keeps streaming while a page comes
back full, then reconciles a stale or short first page by pulling until a
short page arrives — so counts backed by a slow/partial cache still complete.
The first request goes out plain (no offset/limit) so a pre-cached full
result is served when available.
The loader overlay now stays up during a streaming load and reports progress
("Loading results — N loaded").

Search ranking

An exact label match now wins outright, decided before the official-symbol
occurrence heuristic and compared against the main label part — e.g. neuron
ranks above gnathal ganglion neuron.
Real phrase boost (pf: label^250 synonym^120, ps: 0, replacing the no-op
pf: "true"), so short exact multi-word labels such as adult neuron are
retrieved and ranked first instead of falling past the result window.

Thumbnails

Query-result thumbnails are forced to https at render, fixing blank
(mixed-content-blocked) thumbnails served with http:// source URLs.

Results table (VirtualFlyBrain/griddle fork)

VFB now maintains its own griddle fork so patches can land:

Fixes the "Maximum update depth exceeded" crash when typing in the results
filter while a large set is still streaming.
The text filter now re-applies to rows that arrive after you start
typing, so late-streamed matches are not dropped.
The results scroll container reapplies its height on window resize and on
tab/return, so rows are no longer hidden after switching away and back.

Full Changelog: v2.2.8.1...v2.2.8.2