fix: persist all found candidates and guard duplicate adapter names

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

worker/lyra_worker/pipeline.py

@@ -1,8 +1,10 @@
+from dataclasses import replace
 from typing import Sequence
 
 import psycopg
 
 from lyra_worker.adapters.base import SourceAdapter
+from lyra_worker.confidence import score_confidence
 from lyra_worker.quality import quality_class
 from lyra_worker.ranker import rank_candidates
 from lyra_worker.types import Candidate, MBTarget
@@ -103,6 +105,10 @@ def run_pipeline(
     dest_root: str = "/music",
 ) -> None:
     """Real staged acquisition using source-agnostic adapters. Fakes in this plan."""
+    names = [a.name for a in adapters]
+    if len(names) != len(set(names)):
+        raise ValueError(f"adapter names must be unique, got {names}")
+
     # 1. intake
     _set_state(conn, job_id, "matching", "intake")
     target = _load_target(conn, job_id)
@@ -116,16 +122,17 @@ def run_pipeline(
         conn.commit()
         return
 
-    # 2. match
+    # 2. match — search all adapters; score + persist EVERY candidate found
     _set_state(conn, job_id, "matching", "match")
-    candidates: list[Candidate] = []
+    found: list[Candidate] = []
     for adapter in adapters:
-        candidates.extend(adapter.search(target))
+        for c in adapter.search(target):
+            found.append(replace(c, confidence=score_confidence(target, c)))
+    _persist_candidates(conn, job_id, found)
 
     # 3. rank
     _set_state(conn, job_id, "matched", "rank")
-    ranked = rank_candidates(target, candidates, min_confidence)
-    _persist_candidates(conn, job_id, ranked)
+    ranked = rank_candidates(target, found, min_confidence)
     if not ranked:
         _fail(conn, job_id, "no candidate above confidence threshold")
         return