api: implement ranked fixes and grow the round-trip corpus baseline

[webern]

Sub-task of #208. See the parent tracking issue for the full plan and context.

Scope: implement the ranked fixes one at a time; after each, re-run discovery, confirm targeted files flip to PASS with no regressions, append their paths to src/private/mxtest/api/roundtrip-baseline.txt, and commit.

[webern]

Ranked round-trip fix worklist (from PR #225)

Measured over 550 reorder-free candidate files (make dump-api-roundtrip && make classify-api-roundtrip → build/api/classified.json). Ordered by the greedy batch plan (minimal changes → most files). The first two issues land ~26 files outright; all 15 underlying fixes land 71.

#	fix	key signatures	candidate files
#226	stop injecting <identification>/<encoding>	add:encoding (535), add:identification (420)	~535
#227	preserve part-name/part-abbreviation @print-object	attr:part-name@print-object (532)	~532
#228	stop injecting implied defaults (note type, clef line, voice, staff)	add:type (233), add:voice (29), add:line (27)	~280
#230	keep modeled attributes (clef @Number, repeat @times, cancel @location, credit-words @justify)	attr:clef@number (29)	~35
#229	keep zero-valued fields (clef-octave-change, diatonic, stem @default-y)	drop:clef-octave-change (14)	~20
#231	keep accidental editorial attrs (cautionary/editorial/parentheses)	attr:accidental@editorial (5)	~9

Distance-1 instant wins (one fix → PASS): attr:stem@default-y (#229), drop:diatonic (#229), and three add:encoding-only files (#226).

Already tracked separately, not re-filed: element drops — staff/voice/lyric/text/etc. (#219), encoding-child reorder (#220), footnote/level (#221). Reorders (278 files) are deferred to divergence policy (#214).

---

Generated by Claude Code