feat: preserve source voice names in mx::api

[webern]

Human Summary

Allows voices to be labeled instead of numbered.

Summary

mx::api keys the voices of a staff by index, and the writer named every voice from that index. A voice whose <voice> element said something else lost its name. MusicXML does not require a voice to be named with a number at all -- <voice>FirstVoice</voice> is legal, and some programs write it -- and a part whose voices are numbered straight through its staves rather than restarting on each one does not survive renumbering either.

VoiceData gains std::string label. Empty, the default, numbers the voice from its index exactly as before. The reader fills it in from the <voice> element and then drops it again wherever it is just the number the writer would have used anyway, so the common case comes back empty and the field stays out of the way.

The name belongs on the voice, not on each note: containment is what states voice membership in this api, and a per-note name would let notes in one voice disagree about which voice they are in.

Which notes get a <voice> element does not change, so a source that leaves it off a chord member (Sibelius does this) still does. The one addition is a voice where no note carried a number -- an authored score, or a voice named with a word -- since there the label is the only thing that can ask for the element.

Non-breaking: one new field defaulting to the behavior mx had before. NoteData::userRequestedVoiceNumber is untouched.

Round-trip corpus

Adds 2 files to roundtrip-baseline.txt (405 -> 407): musuite/testChordNoVoice.xml and musuite/testStringVoiceName.xml.

musuite/testVoiceMapper1.xml, the third file the classifier listed under value:voice, is not fixed here, and this is not the change that will fix it. That file writes voices 4, 5, 4 back to back with no <backup> between them. mx's reader buckets voices by counting backups, so all three notes land in one api voice, and one api voice can only carry one name. Splitting them into separate voices would be correct but would make the writer emit <forward>/<backup> between them, which the source does not have. Getting that file to round-trip means changing how the reader buckets voices and how the writer interleaves them -- a separate piece of work, worth its own issue.

Testing

• New VoiceLabelApiTest: unlabeled voices numbered from their index, word names round-tripped, a lone voice named 5 written as 5 and stable on a second write, and a name that matches the index dropped on read (*VoiceLabel*: 18 assertions in 4 test cases)
• make api-test (6332 assertions in 586 test cases)
• make api-roundtrip (407 passed, 0 failed of 407 pinned)
• make api-roundtrip-discover reports exactly the 2 new passes and no regressions
• make fmt

References

• Progresses the round-trip corpus work tracked in api: expand corpus coverage of the api round-trip pass-list #208
• Stacked on feat: round-trip a redundant staves element in mx::api #421, feat: support the direction directive attribute in mx::api #420, feat: expose the identification source in mx::api #419, feat: model note playback velocity in mx::api #417. The base branch is feat: round-trip a redundant staves element in mx::api #421's branch, so this diff shows only the voice change.