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/explore/ranker_test.go

@@ -0,0 +1,102 @@
+package explore
+
+import (
+	"bytes"
+	"math"
+	"testing"
+)
+
+func TestDefaultModelScoreOrdering(t *testing.T) {
+	m := DefaultModel()
+
+	// A popular, in-library exact match must outscore an obscure
+	// substring match.
+	strong := RankFeatures{NameMatch: 1.0, LogPopularity: 0.9, InLibrary: 1.0}
+	weak := RankFeatures{NameMatch: 0.2, LogPopularity: 0.1}
+
+	if m.Score(strong) <= m.Score(weak) {
+		t.Errorf("strong candidate %.3f should outscore weak %.3f",
+			m.Score(strong), m.Score(weak))
+	}
+}
+
+func TestUpdateMovesTowardLabel(t *testing.T) {
+	m := DefaultModel()
+
+	// A candidate the user repeatedly clicks should see its predicted
+	// probability rise after training on positive labels.
+	f := RankFeatures{NameMatch: 0.4, LogPopularity: 0.2}
+
+	before := m.Probability(f)
+
+	for range 50 {
+		m.Update(Sample{Features: f, Label: 1.0}, 0.1)
+	}
+
+	after := m.Probability(f)
+
+	if after <= before {
+		t.Errorf("probability should rise toward positive label: before=%.4f after=%.4f",
+			before, after)
+	}
+}
+
+func TestUpdateLearnsNegative(t *testing.T) {
+	m := DefaultModel()
+
+	f := RankFeatures{NameMatch: 0.9, LogPopularity: 0.9}
+
+	before := m.Probability(f)
+
+	// Shown repeatedly, never clicked — probability should fall.
+	for range 50 {
+		m.Update(Sample{Features: f, Label: 0.0}, 0.1)
+	}
+
+	after := m.Probability(f)
+
+	if after >= before {
+		t.Errorf("probability should fall toward negative label: before=%.4f after=%.4f",
+			before, after)
+	}
+}
+
+func TestMatchStrengthTiers(t *testing.T) {
+	if matchStrength(true, false, false, false) != 1.0 {
+		t.Error("exact match should be 1.0")
+	}
+
+	if matchStrength(false, false, false, false) != 0.0 {
+		t.Error("no match should be 0.0")
+	}
+
+	// Tiers must be strictly ordered.
+	exact := matchStrength(true, false, false, false)
+	prefix := matchStrength(false, true, false, false)
+	word := matchStrength(false, false, true, false)
+	sub := matchStrength(false, false, false, true)
+
+	if !(exact > prefix && prefix > word && word > sub) {
+		t.Errorf("tiers not strictly ordered: %v %v %v %v", exact, prefix, word, sub)
+	}
+}
+
+func TestModelRoundTrip(t *testing.T) {
+	m := DefaultModel()
+	m.Bias = 0.123
+	m.Weights[0] = 0.777
+
+	var buf bytes.Buffer
+	if err := SaveModel(&buf, m); err != nil {
+		t.Fatalf("SaveModel: %v", err)
+	}
+
+	got, err := LoadModel(&buf)
+	if err != nil {
+		t.Fatalf("LoadModel: %v", err)
+	}
+
+	if math.Abs(got.Bias-m.Bias) > 1e-9 || math.Abs(got.Weights[0]-m.Weights[0]) > 1e-9 {
+		t.Errorf("round trip mismatch: got %+v want %+v", got, m)
+	}
+}