feat: tuplet numbering and portion fidelity

[webern]

Human Summary

Adds support for nested tuplets and for controlling more about how a tuplet is spelled in the MusicXML. I'm not entirely sure if the LLM understand how nested tuplets work because I don't know well enough to check the synthetic file it created; but the unblocked wild corpus files more or less prove that users can do what they need to do now.

Summary

Two related pieces of tuplet work, plus three attributes that had no home in mx::api.

Numbering. TupletStart and TupletStop carry a SpannerNumber instead of a raw int numberLevel, and tuplets join the other spanner families in SpannerResolver. Tuplets were the
last spanner family whose numbers the caller had to allocate by hand, which is exactly what
nested tuplets need. An author who gives both ends of a tuplet the same identity now gets the
numbers assigned for them: nested tuplets overlap in the serialized stream and are handed 1 and
2, while tuplets that follow one another both get 1. Two unnumbered tuplets open at once are
reported through Diagnostics, since a reader defaults both to 1 and cannot tell them apart.

Portion fidelity. A <tuplet> may spell its ratio out in <tuplet-actual> and
<tuplet-normal> or leave both out and let the note's <time-modification> speak for it. The
reader accepted both shapes and threw the difference away; the writer then put both portions on
every start, turning an unspecified note value into a <tuplet-type>maxima</tuplet-type> the
source never had. That was the sole remaining round-trip blocker on seventeen wild corpus files.

The reader now records which shape the file used, and the writer states a portion only where it
says something the note does not already say. That rule is what nested tuplets need: an inner
triplet of eighths inside a triplet of quarters gives its notes a cumulative 9:4, so the inner
tuplet's own 3:2 differs from the note and is written, while the outer tuplet repeats its note's
3:2 and is left out. TupletStart::writeActual and writeNormal are fidelity fields in the
ClefData::writeStaffNumber style: unspecified applies the rule, yes/no force it either way, so
a caller who finds the portions noisy can turn them off. Inside a portion, only the parts that
were stated are written.

New attributes. show-type (two Bools, modeled like show-number), line-shape
(TupletLineShape), and bracket on a tuplet stop.

Breaking for callers that set numberLevel: number = SpannerNumber{level} to write it and
number.level() to read it back. Callers that relied on both portions always being written need
writeActual/writeNormal set to yes.

Testing

• make api-test (6078 assertions in 698 test cases), including new Tuplet and
  TupletNumber suites covering the reader, the writer's rule, the fidelity fields, the
  three attributes, identity and explicit numbering, and both diagnostics
• make core-roundtrip-test with the pinned count 840 -> 841
• make api-roundtrip 435 of 435 pinned, up from 415
• make fmt-check

Twenty files join roundtrip-baseline.txt: seventeen wild
(foundsuite/Invention_5, ksuite/k011a_Tuplets, lysuite/ly23a, ly23b, ly23c, ly23d,
mjbsuite/PreserveTimeModificationNormalType, the nine musuite tuplet and duration-rounding
files, rpatters1/tuplet_singletons-ref) and three synthetic, one of which is the new
data/synthetic/tuplet-nested.3.0.xml fixture. ly23d_Tuplets_Nested.xml is the corpus's one
genuinely nested tuplet.

References

• Progresses Consider a tuplet authoring helper. #434. The authoring helper it asks for is still open; this removes the hand-allocated
  numbers that made nested tuplets awkward to write.
• Follows NoteWriter infers <normal-type> from sibling notes instead of using DurationData::timeModificationNormalType #428: <normal-type> is the author's statement, and the same reasoning now applies to
  the tuplet portions.
• Related to NotationsWriter emits all spanner stops before all starts, breaking single-note spanners #429: a tuplet that starts and stops on one note is still written start-first, now
  paired by SpannerNumber rather than by number level.
• Related to Support spanners specified as a single object instead of start and stop pairs. #393.

[webern]

I can't tell from this how I'm supposed to use it or when I can leave it unspecified. Please use /mx-comments and write this for a human user that does not have your session context. Explain what it is, what it does, when it's required and when it's optional.

[webern]

I can't tell from this how I'm supposed to use it or when I can leave it unspecified. Please use /mx-comments and write this for a human user that does not have your session context. Explain what it is, what it does, when it's required and when it's optional.

[webern]

The introduction of "inner tuplet" without more context is bad. "Inner tuplet" is not a common use case so you have to explain "In the case of nested tuplets..."