feat: autotag scoring overhaul, dump-based explore index, and lyrics search

Consolidates in-progress work across autotag, explore, and library:

- autotag: beets/Picard-informed scoring engine — ID-first matching, VA
  handling, recommendation tiers, and a merged distance/rank cascade, with
  an eval harness for regression tracking.
- explore: offline MusicBrainz dump import/incremental refresh replaces the
  legacy tier crawl; index-first local search with fuzzy matching and a
  dedicated ranker; disk-free guards for dump downloads.
- library: artist-credit extraction and matching.
- lyrics: owned-library lyric search (FTS) with LRCLIB backfill.

Also: rewrite README to be user-focused, and migrate upstream to
git.ljones.me/yonlu/yellowjacket.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/app.go

@@ -247,20 +247,28 @@ func (yj *YellowJacketApp) OnStartup(ctx context.Context) {
 		RepopulatePlaylists: yj.playlist.RepopulateFromM3U,
 		ResolvePhantoms:     yj.playlist.ResolvePhantomTracksAfterScan,
 		OnAllScansComplete: func() {
-			// Index new library artists (blocks until done).
-			yj.explore.IndexNewArtists()
-			yj.explore.WaitForIndexIdle()
-
-			// Populate local_*_id cross-reference columns on
-			// explore_index so "is this in my library?" is O(1).
+			// Fold the local library into the search index: every
+			// MB-verified owned artist/album/track is upserted and
+			// flagged in_library, straight from the library tables
+			// with no API calls.  Deep discographies stay lazy.
 			yj.explore.PopulateLocalCrossReferences()
 
-			// Always start the full build — it's incremental and
-			// will skip tiers that are already fresh.  This ensures
-			// sitewide + similar artist tiers run even if the index
-			// already has library data.
+			// Start (or resume) the dump-based index build.  Skips
+			// itself once the one-time import has completed, so this
+			// is cheap on every startup.
 			yj.explore.StartIndexBuild()
 
+			// Fold in any new incremental listen dumps to keep
+			// popularity fresh (weekly-gated, background, no API).
+			// No-op while the full import above is still running.
+			yj.explore.RefreshListenCounts()
+
+			// Refresh the lyric-search FTS index from the just-scanned
+			// library, then backfill any missing lyrics from LRCLIB in
+			// the background (bounded, resumable, idempotent).
+			yj.explore.RebuildLyricsIndex()
+			yj.explore.BackfillLibraryLyrics()
+
 			// Sweep the autotag queue for newly-discovered pending
 			// items so the user sees match scores ready when they
 			// next open the review page.  The worker is idempotent
@@ -276,6 +284,10 @@ func (yj *YellowJacketApp) OnStartup(ctx context.Context) {
 	yj.library.SetRemovalHooks(library.RemovalHooks{
 		StopPlayback: func() { yj.player.UnloadTrack() },
 		CompactQueue: yj.queue.CompactAfterLibraryRemoval,
+		// Removal deletes owned content outside a scan, so force the
+		// gated library-sync steps to re-run on the next launch and
+		// clear stale in_library flags / orphaned lyric-index rows.
+		PostRemove: yj.explore.InvalidateLibrarySync,
 	})
 
 	// Register playback finished handler to drive queue auto-advance.
@@ -333,6 +345,21 @@ func (yj *YellowJacketApp) OnStartup(ctx context.Context) {
 func (yj *YellowJacketApp) OnBeforeClose(ctx context.Context) bool {
 	w, h := wailsruntime.WindowGetSize(ctx)
 
+	// Guard against a bogus size clobbering a good saved one.  During
+	// teardown / hot-reload the runtime can report a zero or below-
+	// minimum size; persisting that would shrink the window to the
+	// minimum on next launch.  Keep the previously-saved size instead.
+	if w < config.MinWidth || h < config.MinHeight {
+		yj.logger.Warn("OnBeforeClose: ignoring bogus window size",
+			"width", w,
+			"height", h,
+			"kept_width", yj.appConfig.Window.Width,
+			"kept_height", yj.appConfig.Window.Height,
+		)
+
+		return false
+	}
+
 	yj.logger.Info("OnBeforeClose: saving window state",
 		"width", w,
 		"height", h,
@@ -393,7 +420,24 @@ func (yj *YellowJacketApp) OnDomReady(ctx context.Context) {
 		// index build directly.  If scans WERE queued, the
 		// OnAllScansComplete hook starts it after they finish.
 		if yj.library.GetScanQueueLength() == 0 && !yj.library.IsScanActive() {
+			// Keep the search index's in_library flags and owned-entity
+			// rows in sync.  Gated: the library is unchanged here, so
+			// this only does work on the first launch after an upgrade
+			// or index wipe — steady-state launches skip the write burst.
+			yj.explore.PopulateLocalCrossReferencesIfNeeded()
+
 			yj.explore.StartIndexBuild()
+
+			// Weekly-gated incremental popularity refresh (background,
+			// no API). No-op while the full import is running.
+			yj.explore.RefreshListenCounts()
+
+			// Same for the lyric-search index.  The backfill keeps it in
+			// sync incrementally, so on an unchanged library the full
+			// rebuild is redundant and gated out; the backfill still runs
+			// to fill any remaining gaps.
+			yj.explore.RebuildLyricsIndexIfNeeded()
+			yj.explore.BackfillLibraryLyrics()
 		}
 
 		// Kick off the autotag prefetch worker so any unscored