CMRT 27: Pipeline: Wire MBID -> SimHash -> overlap -> quality -> group assignment

[Swarnadip-Kar]

Wires CMRT 24-26 together into something that actually runs: a new
CmrtGroupingService that AcoustIdWorker calls after every AcoustID lookup,
plus the two DAO methods it turned out to need that didn't exist yet.

CmrtGroupingService::processTrack(trackId, musicbrainzRecordingId)

1. Load fingerprint_metadata for trackId; no-op if missing/invalid.
2. Score quality via QualityScorer.
3. Candidate groups: getGroupsForMbid() if an MBID was found, else
   getAllGroupCandidates() as a fallback.
4. Per candidate: SimHash pre-filter, then (only if that passes) full
   FingerprintMatcher::compare() against the candidate's canonical .chroma.
5. Best match across all candidates that cleared the threshold (not just
   the first one): compare quality scores. Higher score -> replaceCanonical();
   otherwise -> assignToExistingGroup() as a regular member.
6. No match: createNewGroup() -- this track is the first seen for this
   mastering.

New DAO methods (TrackFingerprintDao)

• getTrackQualityInfo(trackId) -> std::optional
• updateCanonicalTrack(groupId, newCanonicalTrackId)
• updateMemberOffset(trackId, offsetFromCanonical)

The latter two didn't exist before this PR -- addCmrtMember() only inserts,
and nothing previously updated canonical_track_id outside
clearFingerprintData()'s own inline SQL.

Wiring into AcoustIdWorker

m_pGroupingService is constructed right after m_pFingerprintDao in doRun()
(and reset right before it, in the reverse order) since it holds a
reference into that DAO, not ownership. processJob() calls
m_pGroupingService->processTrack() after the match-found and
no-match-found outcomes, gated behind kCmrtAutoGroupingEnabledConfigKey
(new preference key, defaults to true).

Notes for reviewers

• getTrackQualityInfo() returns std::optional, not this DAO's usual
  std::unique_ptr -- flagged in case the inconsistency within
  TrackFingerprintDao itself is unwanted; happy to change it.
• The cache-hit path in processJob() (Step 2) does not call
  processTrack() yet -- left as a TODO(XXX) with reasoning in the comment,
  not silently dropped.
• replaceCanonical()'s "recalculate everyone else's offset" loop reads one
  .chroma file per remaining group member. Fine for expected group sizes;
  flagged as a possible future hotspot, not optimized preemptively.

Depends on: cmrt-matcher-core (#35)

[coveralls]

Coverage Report for CI Build 27872458852

Coverage decreased (-0.1%) to 30.317%

Details

• Coverage decreased (-0.1%) from the base build.
• Patch coverage: 239 uncovered changes across 3 files (0 of 239 lines covered, 0.0%).
• 2 coverage regressions across 1 file.

Uncovered Changes

File	Changed	Covered	%
src/musicbrainz/cmrtgroupingservice.cpp	165	0	0.0%
src/library/dao/trackfingerprintdao.cpp	68	0	0.0%
src/musicbrainz/acoustidworker.cpp	6	0	0.0%

Coverage Regressions

2 previously-covered lines in 1 file lost coverage.

File	Lines Losing Coverage	Coverage
src/library/dao/trackfingerprintdao.cpp	2	0.81%

---

Coverage Stats

Relevant Lines:	116694
Covered Lines:	35378
Line Coverage:	30.32%
Coverage Strength:	59035.14 hits per line

---

💛 - Coveralls