Graph quality: fix 7 deselected CI tests before v5.0.0 release

[Simon-McIntosh]

Context

The graph-quality CI job in docker-build-push.yml gates the release pipeline. 7 tests are currently deselected (not continue-on-error) because the IMAS-only graph data pushed to GHCR has known issues. These must be resolved before the final v5.0.0 release.

Failing Tests & Resolution Paths

1. test_facility_nodes_exist (IMAS-only structural)

Error: No Facility nodes in graph
Root cause: IMAS-only graph has no Facility nodes by design.
Fix: Skip this test when running against IMAS-only graph. Add a @pytest.mark.skipif that checks for GraphMeta.facilities == [] or add a --imas-only pytest flag.

2. test_unit_nodes_have_relationships (361 orphaned Unit nodes)

Error: 361 Unit nodes are orphaned (no HAS_UNIT)
Root cause: Unit nodes created during DD build but not linked to IMASNode via HAS_UNIT relationship.
Fix: In the DD build pipeline (imas_codex/graph/dd/), ensure every Unit node created is linked to at least one IMASNode. Rebuild and re-push the graph.

3. test_cocos_reference_nodes (0 COCOS nodes, expected 16)

Error: assert 0 == 16
Root cause: COCOS reference data not built into the IMAS-only graph package.
Fix: Include COCOS build step in the IMAS DD build pipeline. The COCOS reference nodes define sign conventions for magnetic field quantities. Rebuild and re-push.

4. test_dd_versions_linked_to_cocos (0 DDVersions linked)

Error: Expected >=17 DDVersions linked to COCOS, got 0
Root cause: Same as #3 — no COCOS data in graph.
Fix: Same as #3.

5. test_all_labels_in_schema (EmbeddingChange undeclared)

Error: Graph contains labels not in schema: {'EmbeddingChange'}
Root cause: EmbeddingChange nodes exist in the graph but are not declared in schemas/imas_dd.yaml.
Fix: Either add EmbeddingChange class to the LinkML schema, or clean up these nodes from the graph before pushing.

6. test_no_undeclared_properties (cluster_input_hash, test_case, test_prop)

Error: Undeclared properties on DDVersion, IMASNode, IMASSemanticCluster
Root cause: Properties written to graph during development/testing but not declared in schemas.
Fix:

• DDVersion.cluster_input_hash → add to schemas/imas_dd.yaml (legitimate property for cluster build tracking)
• IMASNode.test_case, IMASNode.test_prop → remove from graph (test artifacts)
• IMASSemanticCluster.tags → add to schema or remove from graph

7. test_enum_values_valid (invalid physics_domain values)

Error: Invalid enum values including plasma_profiles, summary, core_profiles, etc.
Root cause: IMASNode physics_domain values don't match PhysicsDomain enum in schema. Old/stale values remain from earlier DD builds.
Fix: Run migration to normalize physics_domain values to match the enum. Invalid values should be mapped to valid enum members or cleared.

Execution Plan

1. Schema fixes (tag image for ACR deployment #5, move main run-server cli within package #6): Add EmbeddingChange and cluster_input_hash to LinkML schemas → uv run build-models --force
2. Graph migration (deploy containers #2, move main run-server cli within package #6, propagate ids_set amongst tools #7): Run Cypher to fix orphaned units, remove test artifacts, normalize physics_domain values
3. COCOS build (Main #3, Server rewrite. Focused semantic search with cached embedings. #4): Ensure COCOS data is included in DD build pipeline
4. Test skip (stdio and streamable-http containers #1): Add conditional skip for IMAS-only graph structural test
5. Re-push graph: uv run imas-codex graph push --dev for imas-codex-graph-imas
6. Remove deselects: Update CI workflow to run all tests
7. Verify: Full CI pass with no deselects or continue-on-error

Priority

Blocking for v5.0.0 final release (not blocking RC1).