[mjc]

Depends on #3670. Once that merges, this PR will show only the lookup changes.

Sonarr and Radarr sync load movie and episode rows before processing the response, then query some of the same values again for each record. Reuse those rows for previous file IDs and paths, movie and episode monitored state, and the series audio-language fallback.

Also remove repeated work in the same loops:

• Load episodes eligible for subtitle searches once per series, after subtitle indexing, using the existing exclusion and retry rules.
• Reuse filesystem size checks when the response reports an undersized file. Keep the reported size as the stored value.
• Index tags and language profiles once per sync. Preserve tag order, duplicate labels, and profile priority.

Movie updates still happen as each record is processed. Bulk updates require the file ID and path to match the loaded values; if another update has replaced the file, the stale write is skipped so the newer file and subtitle index are kept. Monitored-state checks use the loaded rows; changes made by another sync are picked up on the next run. Tag and STRM settings are read during processing.

Measurements

These SQLite fixtures contain 205 records with changed titles. They run the sync functions and parsers with HTTP, subtitle indexing and events stubbed. Both versions perform 205 updates in each changed-record fixture and no updates when every movie is already unmonitored. Stored values match, excluding timestamps. The subtitle-search fixture also queues the same episode IDs.

Sync	SELECTs before	SELECTs after
Movies	206	1
Monitored-only movies, all monitored	411	1
Monitored-only movies, all already unmonitored	206	1
Monitored-only movies, all changing to unmonitored	411	1
Episodes, searches deferred	207	2
Episodes needing the series language fallback, searches deferred	412	3
Episodes with due missing subtitles, searches enabled	617	4

For 205 undersized files, filesystem size checks fall from 410 to 205 in both movie and episode sync.

The tag/profile comparison parses 1,000 records against 500 tags and 100 profiles, with three tags per record. It includes building the indexes and uses the legacy audio-profile fallback. Median time across five runs is listed below; parsed values match.

Parser	Before	After
Movies	191 ms	156 ms
Series	143 ms	130 ms

Tests

Tests check query counts as the library grows, null audio-language fallbacks and fresh state on the next sync, monitored-state refresh between syncs, new unmonitored movies, duplicate movie records, changed-file indexing, insert collisions, and cancellation after completed writes. Search tests cover retry windows, exclusions, and missing subtitles discovered during indexing. File tests check one filesystem lookup per undersized file, STRM handling, and standalone parsing when a file is absent. Tag/profile tests cover ordering, duplicates, ID matching, and settings changes during processing.

Concurrent replacement tests change the file ID, path, or both after rows are loaded, then check that stale writes and their update events are skipped. Episode tests also check that skipped updates do not schedule subtitle searches.

The sync test fixtures now supply previous-file state and numeric job IDs. Changed production functions have parameter and return annotations.