feat: expose the identification source in mx::api - #419

## Human Summary

Adds support for `directive`.

## Summary

`<direction directive="yes">` aligns a direction with the measure's time
signature rather than with the note it is anchored to. That is how a
tempo or style marking at the head of a movement is usually placed. The
api had no field for it, so the reader discarded it and the attribute
was dropped on write.

`DirectionData` gains `Bool directive`, following the api's tri-state
convention: `unspecified` does not write an attribute, `yes` and `no`
are written verbatim.

`<harmony>` does not carry `directive` even though it shares `placement`
and `system` with `<direction>`, so the reader only looks for it on
`<direction>`.

Non-breaking: one new field defaulting to `unspecified`.

## Round-trip corpus

Adds 3 files to `roundtrip-baseline.txt` (399 -> 402).
`attr:direction@directive` was the third entry on the classifier
worklist.

## Testing

- [x] Two new `DirectionWriter` impl tests: the api -> core -> api round
trip of `yes`, and that `unspecified` leaves the attribute off entirely
(`*directive*`: 9 assertions in 2 test cases)
- [x] `make api-test` (6306 assertions in 579 test cases)
- [x] `make api-roundtrip` (402 passed, 0 failed of 402 pinned)
- [x] `make api-roundtrip-discover` reports exactly the 3 new passes and
no regressions
- [x] `make fmt`

## References

- Progresses the round-trip corpus work tracked in #208
- Stacked on #419, which is stacked on #417. The base branch is #419's
branch, so this diff shows only the directive change.

---------

Signed-off-by: Matthew James Briggs <matthew.james.briggs@gmail.com>

## Human Summary

Not a functional feature, but we now respect a file that emits
`<staves>1</staves>` when we write it back.

## Summary

MusicXML states a part's staff count in `<attributes>/<staves>`, and a
part that omits the element has one staff. mx wrote the element only
when the part had more than one staff, so a source that spelled out the
redundant `<staves>1</staves>` lost it.

`PartData` gains `bool writeStaffCount`, defaulting to false. It only
affects a single-staff part: a part with more than one staff writes its
count either way, since without it every note would land on staff one.
The reader sets it when the source spelled the element out and the count
is one, so the common case stays at the default.

`PartReader::calculateNumStaves` now also reports whether a `<staves>`
element was seen. The count alone cannot tell you, since one is both
"the source said one" and "the source said nothing".

Non-breaking: one new field defaulting to the behavior mx had before.

## Round-trip corpus

Adds 3 files to `roundtrip-baseline.txt` (402 -> 405). `drop:staves` was
the fourth entry on the classifier worklist.

## Testing

- [x] New `StaffCountApiTest` covers the single-staff default, the
single-staff override, and a two-staff part that writes its count
regardless (`*StaffCount*`: 8 assertions in 3 test cases)
- [x] `make api-test` (6314 assertions in 582 test cases)
- [x] `make api-roundtrip` (405 passed, 0 failed of 405 pinned)
- [x] `make api-roundtrip-discover` reports exactly the 3 new passes and
no regressions
- [x] `make fmt`

## References

- Progresses the round-trip corpus work tracked in #208
- Stacked on #420, #419, #417. The base branch is #420's branch, so this
diff shows only the staff-count change.

---
_Generated by [Claude
Code](https://claude.ai/code/session_01RmcR2rHB4dLWjtPWqGrQxV)_

---------

Signed-off-by: Matthew James Briggs <matthew.james.briggs@gmail.com>