feat(enrichment): AcoustID audio-fingerprint identification (opt-in)

Text search can only guess the version; the definitive fix is content-based —
fingerprint the actual audio with Chromaprint (fpcalc) and look it up on
AcoustID, which maps the fingerprint to the EXACT MusicBrainz recording (the
approach Lidarr uses). Sidesteps the studio-vs-live ambiguity entirely.

- lib/acoustid_match.py: pure response parsing + config gating (unit-tested);
  normalizes AcoustID hits into the same candidate shape as mb_match so the
  review UI + editor Match popup render fingerprint and text hits identically.
- server.py: _fpcalc (Chromaprint subprocess), _acoustid_lookup (throttled,
  offline-guarded HTTP), _identify_by_fingerprint (also available to the
  library-enrichment pipeline), and POST /api/enrichment/identify (upload the
  master audio → candidates).
- Fully OPT-IN and graceful: absent the fpcalc binary or an ACOUSTID_API_KEY
  the whole path is a no-op / 503 and the text matcher runs unchanged.

Requires (both optional): the `fpcalc` (Chromaprint) binary on PATH/$FPCALC,
and a free AcoustID application key in $ACOUSTID_API_KEY. Pure parsing/gating
is unit-tested; the fpcalc + live-lookup path needs those two to exercise.

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

lib/acoustid_match.py

@@ -0,0 +1,128 @@
+"""AcoustID audio-fingerprint identification for MusicBrainz enrichment.
+
+A flat MusicBrainz *text* search ties every take of a song at the same score —
+studio, a dozen live bootlegs, and every compilation — so "AC/DC — Highway to
+Hell" returns junk (see lib/mb_match.py's canonical re-ranking, which mitigates
+it). The definitive fix is content-based: fingerprint the actual audio with
+Chromaprint (`fpcalc`) and look it up on AcoustID, which maps the fingerprint
+straight to the *exact* MusicBrainz recording — the same approach Lidarr uses.
+
+This module is the PURE half (no network, no subprocess): response parsing +
+config gating, so it is unit-testable in isolation. server.py owns the `fpcalc`
+subprocess and the throttled HTTP GET to api.acoustid.org.
+
+Operational requirements (both optional — absent ⇒ this path is a graceful
+no-op and the text matcher still runs):
+  * `fpcalc` (Chromaprint) on PATH or at $FPCALC — generates the fingerprint.
+  * an AcoustID application API key in $ACOUSTID_API_KEY — free from
+    https://acoustid.org/new-application ; AcoustID etiquette limits to ~3 req/s.
+"""
+
+import os
+
+ACOUSTID_API_ROOT = "https://api.acoustid.org/v2"
+
+# The `meta` fields we ask AcoustID to return so a hit resolves to displayable
+# metadata without a second MusicBrainz round-trip.
+LOOKUP_META = "recordings+releasegroups+compress"
+
+# Mirror mb_match._SECONDARY_SKIP: release-group secondary types that mark a
+# non-canonical (live/comp/remix) release, so we can flag the studio take.
+_SECONDARY_SKIP = {
+    "live", "compilation", "remix", "dj-mix", "mixtape/street",
+    "demo", "interview", "audiobook", "spokenword",
+}
+
+
+def api_key(explicit: str | None = None) -> str:
+    """The AcoustID application API key: an explicit value (e.g. a host setting)
+    wins, else $ACOUSTID_API_KEY, else "" (⇒ fingerprinting disabled)."""
+    return (explicit or os.environ.get("ACOUSTID_API_KEY") or "").strip()
+
+
+def is_configured(explicit_key: str | None = None) -> bool:
+    """True when an API key is available. `fpcalc` presence is checked by
+    server.py (it owns the binary lookup); both are required to actually run."""
+    return bool(api_key(explicit_key))
+
+
+def _rg_is_studio(rg: dict) -> bool:
+    if str(rg.get("type", "")).lower() != "album":
+        return False
+    secs = {str(s).lower() for s in (rg.get("secondarytypes") or [])}
+    return not (secs & _SECONDARY_SKIP)
+
+
+def _best_group(recording: dict) -> dict:
+    """Prefer a studio Album release-group for the display album, else the first."""
+    groups = [g for g in (recording.get("releasegroups") or []) if isinstance(g, dict)]
+    if not groups:
+        return {}
+    groups = sorted(groups, key=lambda g: 0 if _rg_is_studio(g) else 1)
+    return groups[0]
+
+
+def _first_artist(recording: dict) -> str:
+    for a in (recording.get("artists") or []):
+        if isinstance(a, dict) and a.get("name"):
+            return str(a["name"])
+    return ""
+
+
+def parse_lookup_response(body: dict) -> list[dict]:
+    """Normalize an AcoustID /v2/lookup response into the same flat candidate
+    shape as mb_match (recording_id / title / artist / album / year / duration /
+    studio / mb_score / score), so the review UI and the editor's Match popup
+    render fingerprint hits and text hits identically. `mb_score` carries the
+    AcoustID confidence (0-100) — a fingerprint hit is high-signal by nature."""
+    if not isinstance(body, dict) or body.get("status") != "ok":
+        return []
+    out: list[dict] = []
+    seen: set[str] = set()
+    for result in (body.get("results") or []):
+        if not isinstance(result, dict):
+            continue
+        try:
+            score = float(result.get("score") or 0.0)
+        except (TypeError, ValueError):
+            score = 0.0
+        for rec in (result.get("recordings") or []):
+            if not isinstance(rec, dict) or not rec.get("id"):
+                continue
+            rid = str(rec["id"])
+            if rid in seen:
+                continue
+            seen.add(rid)
+            rg = _best_group(rec)
+            year = ""
+            for rel in (rg.get("releases") or []):
+                d = (rel or {}).get("date") or {}
+                y = d.get("year") if isinstance(d, dict) else None
+                if y:
+                    year = str(y)[:4]
+                    break
+            dur = rec.get("duration")
+            try:
+                duration = int(round(float(dur))) if dur else None
+            except (TypeError, ValueError):
+                duration = None
+            out.append({
+                "recording_id": rid,
+                "title": str(rec.get("title", "") or ""),
+                "artist": _first_artist(rec),
+                "album": str(rg.get("title", "") or ""),
+                "year": year,
+                "duration": duration,
+                "isrc": "",
+                "genres": [],
+                "studio": _rg_is_studio(rg),
+                "acoustid_score": round(score, 4),
+                # Fingerprint hits are content-verified, not text-guessed — carry
+                # the AcoustID confidence as the display score band.
+                "mb_score": int(round(score * 100)),
+                "score": round(score, 4),
+                "source": "acoustid",
+            })
+    # Best AcoustID confidence first; studio take breaks ties.
+    out.sort(key=lambda c: (c["acoustid_score"], 1 if c["studio"] else 0), reverse=True)
+    return out