fix(holon-viz): emit valid SysML v2 syntax; add real-parser validation (ufo_types::sysml)

[elasticdotventures]

What

Follow-up from the SysML v2 epic's retrospective + issue #195. The
unmerged spike PR #187 (sysml-v2-parser round-trip investigation) already
found that holon-viz's SysmlV2Emitter doesn't produce parseable SysML
v2 at all — this PR fixes both bugs it found and lands a reusable real-parser
validator, rather than leaving the finding as a spike writeup nobody acted
on.

Bugs fixed in SysmlV2Emitter

1. Comment swallows the closing brace. The emitter wrote the closing
   } on the same line as a trailing // id: ..., kind: ... comment —
   // runs to end of line, so the brace was never actually there
   syntactically.
2. block def is SysML v1, not v2. SysML v1 called this construct
   Block; SysML v2 renamed it to part def. block is not a SysML v2
   keyword — parse()/parse_for_editor() reject it outright.

Both confirmed against the real grammar, not asserted from reading the
spec.

New: ufo_types::sysml

A shared SysmlV2Syntax constraint + validate_sysml_v2() function (and
Satisfies<SysmlV2Syntax> impls for str/String), wired to the
sysml-v2-parser crate (pinned =0.54.0, matching the spike's own
recommendation given its pre-1.0/high-churn release cadence). This gives
every crate that generates SysML v2 text (today: holon-viz,
sysml-derive — see the follow-up PR wiring this into #196) one shared,
real-grammar oracle instead of each hand-rolling its own heuristic (the
"no <> characters" check I originally used to validate #196 was exactly
this kind of heuristic).

Important dependency-graph note: sysml-v2-parser pulls in stacker,
which is not wasm32-compatible. ufo-types itself is never compiled to
wasm in this repo, so it's a normal dependency there. holon-viz, however,
is compiled to wasm (holon-viz-wasm) — so ufo-types is added there
only as a dev-dependency, used in a new test
(tests/sysml_v2_roundtrip.rs) that feeds SysmlV2Emitter's own output
through the real parser. This must not become a runtime dependency of
holon-viz.

Verification

• cargo test -p ufo-types: 17/17 pass (4 new, including one that asserts
  block def is rejected and one that asserts the comment/brace bug is
  rejected — both against the real parser).
• cargo test -p holon-viz: all pass, including the 3 new round-trip
  tests (previously these exact same inputs failed to parse — this is the
  concrete "did we close the round-trip" signal PR spike: sysml-v2-parser round-trip against holon-viz emitter #187's writeup asked
  for).
• Updated the two pre-existing tests that asserted on the literal string
  "block def" to expect "part def" instead (one count changed from 2
  to 3, since nodes and containment edges now correctly share one
  keyword instead of an artificial v1/v2-mixed split).
• cargo check --workspace --all-features: clean.