- The tag parser's cache is now cleared between indexing runs, preventing stale data from being used
 - Multi-value tag fields (genres, all artists) are now properly ingested in the tag value cache
 - Cleaning up removed index records now properly handles the case where only a subset of the records for multi-value tags need to be deleted.
 - Synthesizing missing tag values is now done in the tag parser instead of TrackTags, which resolves some issues with multi-value tag callbacks from libtags not being handled properly

These fixes seem to address all the issues with stale index records we were able to repro (including the issues in https://codeberg.org/cool-tech-zone/tangara-fw/issues/191), but if you've got any more cases with consistent repros then lmk!

Co-authored-by: ailurux <ailuruxx@gmail.com>
Reviewed-on: https://codeberg.org/cool-tech-zone/tangara-fw/pulls/193

It turns out the index *id* controls this, which is probably not good!

We still mostly use the singular 'Artist' tag for e.g. displaying a nice
name in the now playing screen, but where a track has an 'ARTISTS=' tag,
we'll split by semicolon and then use the resulting list to populate an
index of tracks by artist

Podcasts aren't tagged consistently enough for other groupings to make
sense, at least until we implement RSS parsing

Initially set based on filepath, or by genre if the filepath doesn't
match one of our presets

This is ostensibly yet another 'prepare for multithreaded updates'
commit, however it does actually save us another 60(!!) odd milliseconds
per track.