- fix(storage): tracks of two albums with the same title (Greatest Hits, Live)
no longer interleave in a listing: the album id is part of the order
- fix(storage): an artist's album_count and GET /albums?artist_id= now mean
the same thing: albums the artist owns plus albums where it is credited on a
playable track; a property test checks it for every artist and library
- fix(catalog): a NUL byte or invalid UTF-8 in q is a 400 instead of a broken
SQL statement and a 500
- fix(api): clamp paging values with math.MaxInt32 so the package also builds
on 32-bit targets
- docs: search ignores case, including accented letters, but not accents
("bjork" does not find "Bjork"); the earlier comments said otherwise
- test: same-title albums across a page boundary, NUL and invalid UTF-8
- feat(api): GET /api/v1/artists, /albums and /tracks with q (case and accent
insensitive substring), limit and offset (default 50, at most 200) and a
total; filters library_id, artist_id and album_id; GET /artists/{id},
/albums/{id} and /tracks/{id}; any signed-in user can read
- feat(catalog): only tracks with a file that is not flagged missing are
visible, so everything shown can be played; a track with several files is
listed once; album artists without credited tracks (compilations) are listed
- feat(storage): results are ordered the same way on SQLite and PostgreSQL
(plain byte order), LIKE wildcards in the search text are escaped, and a
page costs a fixed number of queries: related items and counts are loaded in
batches
- feat(storage): tracks get a title_key computed in Go (migration 0006, with
a startup refresh), like artists and albums
- test: ordering, paging, search with accents and wildcards, library filter,
duplicate encodings, missing files, details, and a query-count guard on
both databases