* fix: improve autosuggest a11y
some progress on #1629
- works in Chrome on NVDA now
- works in Chrome on VoiceOver now
- shorter aria-labels, don't repeat information like "1 of 3", because
it causes the screen reader to speak too frequently, e.g. when the
selected result hasn't changed but the number of results has. Also both
NVDA and VoiceOver already speak this information
- stop doing a fancy fade animation, just show and hide the input
instantly. I worry it confuses screen readers to have the aria-hidden
attribute in there at all
- stop using a single id to identify the active descendant - give
immutable IDs and then update the aria-activedescendant instead. I think
this is what fixed Chrome.
* fix test
* perf: fetch lists from database on page load
follow-up to #1448, I would like for there not to be a flash on the /community page where the lists suddenly load in, but I would still like to avoid the network request if possible. So when the page first loads, we can load the lists from the database and only fetch them from the network if not in the cache.
* add comment
* fix: move blurhash worker operations to before status rendering
* slight refactor
* avoid sending encoded data back and forth
* move cache outside worker
* chore(npm): Install blurhash
* feat(media): Show blurhash
* fix(media/blurhash): Better sensitive video handling
* feat(media): Preference for using blurhash
* chore(utils/blurhash): Add performance marks
* fix(utils/blurhash): Performance marks
* fix(utils/blurhash): Use correct dimension
* refactor(utils/blurhash): Use constant for number of pixels
* refactor(media): Simplify logic for displaying blurhash
* chore(tests/spec): Attempt to adjust sensitivity tests for blurhash
* chore(tests/spec): Update sensitivity tests for blurhash
* chore(tests/spec): Check for sensitive
* fix(media/blurhash): Handle videos
* fix: Video handling
* fix: Videos
* minor refactoring, fix Svelte warning
* fix: Large inline images and videos
* feat(settings): Rename blurhash setting
* refactor: Use toBlob, block media rendering until blurhash ready
* refactor: Move computations to Web Worker
* fix(workers/blurhash): More error handling
* feat(workers/blurhash): Use quick-lru for caching
* fix: Don't create Context2D needlessly
* fix(workers/blurhash): Increase cache size to 100
* fix(workers/blurhash): Don't resolve promise twice
* fix(utils/decode-image): Ignore data URLs
Throws exception which prevents the image from loading.
* perf: only update draggable x/y state at end of drag
This is more intelligent and more performant than using requestIdleCallback willy-nilly. We can just update the store when the user is actually done dragging the button.
* remove console.log
* consistent syntax
