A spec-by-spec audit of the lyrics tests found assertions duplicated across
layers and within files. Trim them so each test owns one concern:
- model/lyrics_embedded_test.go: scope the embedded TTML specs to dispatch +
language defaulting; the deep cue/agent/byte details are owned by
lyrics_ttml_test.go.
- model/lyrics_test.go: drop the no-marker Enhanced LRC spec; the plain-line
(nil cues) path is already covered by the mixed Enhanced/plain LRC spec.
- model/lyricsfile_test.go: merge the two version-marker rejection specs into a
single DescribeTable (same production branch).
- model/metadata/map_mediafile_test.go: drop the >32KB TTML mapping test; the
tag-length cap is covered in metadata_test.go and TTML parsing in the model
lyrics tests.
- model/metadata/metadata_test.go: drop the 60KB lyrics-cap test, subsumed by
the truncation test that pins the exact ~1MB lyrics cap boundary.
No production code changed; coverage is preserved at the correct layer.
ac64ebef79