fix: ListenBrainz similarity is stored relative to the seed's best, and the candidate arms keep tier order (#6177)

LB's similarity scores are counts (16 to 1445 for recordings, to 10,875
for artists), and they reached a scorer that reads similarity as 0 to 1.
One edge added 64 to 5,800 points under Songs-like, so likes, skips and
recency stopped mattering for any candidate with an edge.

- ResolveListenBrainzTrackEdges divides each edge by the seed's best
  score in its whole cached answer. The cache keeps the raw answer.
- The worker stores matched and unmatched artist rows over the seed's
  top artist.
- Migration 0096 rescales the rows already stored.
- The candidate arms hold sim_score bands so tier 1 stays above tier 2:
  lb_similar 0.5 + 0.5 x edge, tag_overlap halved, likes_overlap 0.3.
  Otherwise a seed's weaker LB edges fell below a shared genre tag.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

internal/db/migrations/0096_similarity_relative_scores.down.sql

@@ -0,0 +1,3 @@
+-- The raw scores are not kept for artists; the worker rewrites every seed with
+-- relative scores on its next fetch either way. Nothing to undo.
+SELECT 1;

internal/db/migrations/0096_similarity_relative_scores.up.sql

@@ -0,0 +1,51 @@
+-- #6177: ListenBrainz similarity scores were stored raw. They are counts (16 to
+-- 1445 for recordings, 16 to 10,875 for artists on the operator's library), and
+-- every reader treats similarity as 0 to 1. Rescale the stored rows relative to
+-- each seed's best score, as the worker now writes them.
+--
+-- A seed's best is taken over everything stored for it: the cached LB answer
+-- and its edges for tracks, matched and unmatched rows for artists. A seed
+-- whose best is already at most 1 is left alone, so a second run changes
+-- nothing.
+
+WITH best AS (
+    SELECT seed_id, max(score) AS score
+    FROM (
+        SELECT track_a_id AS seed_id, score FROM track_similarity WHERE source = 'listenbrainz'
+        UNION ALL
+        SELECT seed_track_id, score FROM listenbrainz_similar_recordings
+    ) s
+    GROUP BY seed_id
+)
+UPDATE track_similarity ts
+   SET score = ts.score / b.score
+  FROM best b
+ WHERE ts.source = 'listenbrainz'
+   AND ts.track_a_id = b.seed_id
+   AND b.score > 1;
+
+-- One statement, so both updates divide by the best as it stood before either.
+WITH best AS (
+    SELECT seed_id, max(score) AS score
+    FROM (
+        SELECT artist_a_id AS seed_id, score FROM artist_similarity WHERE source = 'listenbrainz'
+        UNION ALL
+        SELECT seed_artist_id, score FROM artist_similarity_unmatched WHERE source = 'listenbrainz'
+    ) s
+    GROUP BY seed_id
+),
+matched AS (
+    UPDATE artist_similarity a
+       SET score = a.score / b.score
+      FROM best b
+     WHERE a.source = 'listenbrainz'
+       AND a.artist_a_id = b.seed_id
+       AND b.score > 1
+    RETURNING 1
+)
+UPDATE artist_similarity_unmatched u
+   SET score = u.score / b.score
+  FROM best b
+ WHERE u.source = 'listenbrainz'
+   AND u.seed_artist_id = b.seed_id
+   AND b.score > 1;