fix: take an identifier-only mixed-type link from the instance library

[ehennestad]

Fixes the R2026a test failure in run 34695830401: RoundTripTest/testJsonLdRoundTrip(MetadataType=AnatomicalAtlasVersion) errored with OPENMINDS_MATLAB:MetaTypeRegistry:InvalidKey because the serializer was handed an openminds.internal.MixedTypeReference as if it were a node.

Cause

A property that holds several types stored a bare {"@id": ...} as a MixedTypeReference, whatever the identifier named. When it named a controlled instance, Collection.addNode treated it as a controlled instance and made it a node, and saving failed because a reference of unknown type is not a metadata type. The regenerated toolbox from openMetadataInitiative/openMINDS_MATLAB_pipeline#193 exposed this for AnatomicalAtlasVersion: its linked types gained HasControlledInstance, so the round-trip synthesizer now fills them from real library documents, one of which links usageCondition to a licence by @id. The same defect was already recorded in knownRoundTripGap for ParcellationTerminologyVersion and QuantitativeRelationAssessment.

Controlled terms built from a bare identifier, and IRI strings assigned to mixed-type properties, are already taken from the instance library at construction. A struct carrying only a library IRI was the one input that was not.

Change

• MixedTypeSet: an identifier-only link whose identifier is a controlled instance IRI is taken from the library with openminds.instanceFromIRI and checked against the allowed types. Any other identifier stays a MixedTypeReference. So does a library lookup that fails, with a warning carrying the cause; the library does not always hold what an IRI points to, for example an IRI spelling MNIColin27_1998 where the file is MNI-Colin27_1998.
• Serializer.enqueueDocument: a reference is never queued as a document of its own, at any recursion depth. This also fixes Node.serialize() on a node holding a mixed-type reference, which failed the same way at the default recursion depth.
• knownRoundTripGap: ParcellationTerminologyVersion and QuantitativeRelationAssessment are removed, since both now round-trip.
• Tests: library lookup from an identifier-only link, the fallback with its warning, and a reference getting no document of its own at recursion depth 2.

Not changed

• Typed references to library instances still become stub nodes of a collection and are written as {"@type", "@id"} documents. That is round-trip stable and out of scope here, but in the atlas case it is 1339 near-empty documents out of 1365 and deserves its own issue.
• The R2022a job cannot exercise these paths because the synthesizer caches names in a dictionary, which that release lacks; the call fails inside a try/catch and the affected properties are silently skipped. Worth a separate fix so the R2022a run stops passing vacuously.

🤖 Generated with Claude Code

[github-actions]

Test Results (R2022a)

861 tests   857 ✅  5m 3s ⏱️
 27 suites    4 💤
  1 files      0 ❌

Results for commit b5bc40f.

[github-actions]

Test Results (R2026a)

861 tests   859 ✅  4m 30s ⏱️
 27 suites    2 💤
  1 files      0 ❌

Results for commit b5bc40f.

[codecov]

Codecov Report

❌ Patch coverage is 93.33333% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 80.25%. Comparing base (9030b75) to head (b5bc40f).
⚠️ Report is 4 commits behind head on main.

Files with missing lines	Patch %	Lines
code/+openminds/+base/MixedTypeSet.m	92.30%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #196      +/-   ##
==========================================
- Coverage   80.37%   80.25%   -0.13%     
==========================================
  Files         424      424              
  Lines        4403     4436      +33     
==========================================
+ Hits         3539     3560      +21     
- Misses        864      876      +12     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.