End-to-end live test point for openCAESAR SysMLv2 ingestion

[mzargham]

Follows the Flexo live-test reconciliation: spec divergences only
show up against the real thing. Today tests/integration/sysmlv2/
consumes fixtures already in omg-sysml: RDF form. The conversion
step SysMLv2 source → openCAESAR owl-adapter → omg-sysml: RDF is
NOT exercised — meaning we can't catch:

• Pinned openCAESAR version skew producing an RDF shape different
  from what oracle/adapters/sysmlv2/mapping.py expects
• New omg-sysml: properties in v1.x rendering we don't map
• Round-trip-through-conversion loss in the write-back path (SysMLv2 Ingestion Contract §9: pull RDF write-back into v0.1 scope (read-step confidence) #18)

Options for a live test point:

1. Install openCAESAR JVM toolchain in CI and run
   .sysml → .ttl → flexo-rtm read → flexo-rtm emit → .sysml.json → compare to a pinned reference output. Pros: real end-to-end.
   Cons: pulls JVM + Gradle + openCAESAR dependency graph into CI;
   longer cold-start.

2. Pre-cache a converted-RDF fixture from a known openCAESAR
   version + commit hash of the source .sysml, with the conversion
   step documented as a side-shell-script in scripts/. Run only on
   make sysmlv2-refresh-fixtures. Pros: keeps CI fast and JVM-free.
   Cons: refresh becomes manual; doesn't auto-catch upstream version
   skew.

3. Build a flexo-rtm-sysmlv2-converter container bundling a
   pinned openCAESAR version; CI pulls and runs conversion inside it.
   Pros: container caching keeps CI fast; one place to bump version.
   Cons: another artifact to maintain.

SysMLv2 Ingestion Contract §3 should call out the v0.1 standard
test point — today it's silent on end-to-end testing.

Decision needed:

• Which option (1/2/3) for v0.1?
• If (2): which .sysml file is the canonical input? (ADCS prototype
  has one — if not cached it should be reused.)

[mzargham]

Impl-side ticket blocked on this decision: flexo-rtm#5.