fix(music): stop sharing one folder's cover across unrelated tracks, and merge various-artists compilations into one album

Two real, confirmed bugs in a large flat music library:

- enrich_audio.py's sibling-cover fallback assumed one folder is one
  release. A large flat "chart ranking" folder mixing dozens of unrelated
  artists carried several distinct WMP AlbumArt-cache guids (one per
  original album a track was ripped from), and the fallback picked
  whichever one WMP had copied to Folder.jpg — attaching one unrelated
  release's cover to every other track in the folder. Now refuses to pick
  a cover at all once 2+ distinct guids show up, rather than guess.

- music-app.js's groupMusicEntries grouped by artist first, album second,
  so a various-artists compilation (many genuinely different per-track
  artists, one shared album tag, no album-artist tag at all — a real
  ~20-track soundtrack rip has exactly this shape) could never be
  recognized as one release: every track landed alone in its own artist's
  bucket and got folded into a singleton pile. Now detects an album key
  shared across 2+ distinct artist keys and merges those tracks into one
  compilation card under a "Various" heading instead.

Both verified against real, previously-affected files and live in the
browser: the shared wrong cover is gone, and the compilation renders as
one card with all its tracks in order.

2 files changed, 55 insertions, 0 deletions

diff --git a/packages/meshbay-node/src/meshbay_node/indexer/enrich_audio.py b/packages/meshbay-node/src/meshbay_node/indexer/enrich_audio.py
index a93df51..4fa08ef 100644
--- a/packages/meshbay-node/src/meshbay_node/indexer/enrich_audio.py
+++ b/packages/meshbay-node/src/meshbay_node/indexer/enrich_audio.py
@@ -141,6 +141,18 @@ def _extract_cover(mf) -> bytes | None:
 _COVER_STEM_RANK = ("cover", "folder", "front", "albumart")
 _COVER_EXTS = (".jpg", ".jpeg", ".png")
 
+# Windows Media Player's per-folder thumbnail cache: one `AlbumArt_{guid}_*`
+# pair per *release* it ever cached art for in that folder, keyed by a GUID
+# tied to that release — never to a track. Real folder, real GUIDs (module
+# docstring's revision note): a ~500-track flat "chart ranking" rip mixing
+# dozens of unrelated artists carried seven distinct guids here, each an art
+# leftover from one different original album. `Folder.jpg`/`AlbumArtSmall.jpg`
+# are WMP's own copy of just *one* arbitrary one of those for the folder icon
+# — fine when a folder really is one release (one guid, or none), meaningless
+# once two or more show up: there is no way to tell which track it belongs to,
+# so no candidate in the folder can be trusted as "the" cover for any of them.
+_WMP_ALBUMART_GUID_RE = re.compile(r"^albumart_\{([0-9a-f-]{36})\}_(large|small)$", re.IGNORECASE)
+
 
 def _find_sibling_cover(folder: Path) -> Path | None:
     try:
@@ -151,6 +163,11 @@ def _find_sibling_cover(folder: Path) -> Path | None:
     if not candidates:
         return None
 
+    guids = {m.group(1).lower() for p in candidates
+             if (m := _WMP_ALBUMART_GUID_RE.match(p.stem))}
+    if len(guids) >= 2:
+        return None
+
     def rank(p: Path) -> int:
         stem = p.stem.lower()
         for i, name in enumerate(_COVER_STEM_RANK):
diff --git a/packages/meshbay-node/tests/test_enrich_audio.py b/packages/meshbay-node/tests/test_enrich_audio.py
index 1485a1f..d69a097 100644
--- a/packages/meshbay-node/tests/test_enrich_audio.py
+++ b/packages/meshbay-node/tests/test_enrich_audio.py
@@ -12,6 +12,7 @@ from meshbay_node.indexer.enrich_audio import (
     _artist_album_from_ancestors,
     _clean_tag,
     _extract_cover,
+    _find_sibling_cover,
     _musepack_tags_from_object,
     _split_top_level_folder,
 )
@@ -292,6 +293,43 @@ async def test_enricher_uses_a_sibling_cover_file_when_no_embedded_art(tmp_path,
     assert stored == b"\xff\xd8\xff\xe0fake-jpeg-bytes"
 
 
+def test_find_sibling_cover_refuses_a_folder_that_mixes_several_releases(tmp_path):
+    """
+    A real case: dozens of individually-ripped tracks by unrelated artists
+    sitting flat in one folder (an informal "chart ranking" compilation,
+    never a real per-album folder). The ripper left one `AlbumArt_{guid}_*`
+    thumbnail pair *per original release* it pulled a track from, plus its
+    own arbitrary `Folder.jpg` copy of just one of them. Two or more distinct
+    guids among the candidates is direct proof this folder is not a single
+    release, so no image here can be safely called "the" cover — picking
+    `Folder.jpg` anyway (the pre-fix behavior) silently attached one
+    unrelated release's art to every other track in the folder.
+    """
+    folder = tmp_path / "_SomeChartFolder"
+    folder.mkdir()
+    (folder / "AlbumArt_{2C5E5D27-1075-43B9-92EE-351846034C67}_Large.jpg").write_bytes(b"one")
+    (folder / "AlbumArt_{2C5E5D27-1075-43B9-92EE-351846034C67}_Small.jpg").write_bytes(b"one")
+    (folder / "AlbumArt_{427CA566-E7E9-4ECD-9ED1-1DDADB9D07E8}_Large.jpg").write_bytes(b"two")
+    (folder / "AlbumArt_{427CA566-E7E9-4ECD-9ED1-1DDADB9D07E8}_Small.jpg").write_bytes(b"two")
+    (folder / "AlbumArtSmall.jpg").write_bytes(b"whichever")
+    (folder / "Folder.jpg").write_bytes(b"whichever")
+
+    assert _find_sibling_cover(folder) is None
+
+
+def test_find_sibling_cover_still_picks_folder_jpg_for_a_single_release(tmp_path):
+    """Regression guard: a real single-release folder (one guid, or none at
+    all, same shape as the existing sibling-cover test above) must keep
+    picking its Folder.jpg exactly as before this fix."""
+    folder = tmp_path / "Some Artist" / "Some Album"
+    folder.mkdir(parents=True)
+    (folder / "AlbumArt_{2C5E5D27-1075-43B9-92EE-351846034C67}_Large.jpg").write_bytes(b"one")
+    (folder / "AlbumArt_{2C5E5D27-1075-43B9-92EE-351846034C67}_Small.jpg").write_bytes(b"one")
+    (folder / "Folder.jpg").write_bytes(b"one")
+
+    assert _find_sibling_cover(folder) == folder / "Folder.jpg"
+
+
 async def _run(enricher, entry, path, root_path=None):
     done = asyncio.get_event_loop().create_future()