Standardize SSSOM IRI/CURIE handling with curies + sssom-py

[turbomam]

Problem

Current SSSOM consumers/producers in this repo use a mix of ad hoc string heuristics for object_id parsing (hostname checks, OBO-only regexes). This is brittle and diverges from SSSOM-native tooling.

Why this matters

• We already observed mixed vs IRI vs CURIE forms in legacy mapping files.
• Heuristic parsing can silently misclassify prefixes and break downstream analysis.

Proposal

1. Add curies and sssom-py as dependencies for mapping workflows.
2. Build one shared utility for:
   • reading curie_map from SSSOM metadata,
   • expansion/compaction through curies.Converter,
   • strict normalization rules for subject_id/predicate_id/object_id.
3. Replace remaining heuristic parsing in scripts under:
   • metpo/analysis/
   • metpo/presentations/
   • metpo/scripts/
   • metpo/pipeline/
4. Add validation step that fails when unresolved/ambiguous CURIE prefixes remain.

Acceptance criteria

• No script relies on hostname substring matching for prefix classification.
• SSSOM files round-trip via SSSOM tooling without identifier shape regressions.
• Curie map is complete for every CURIE prefix emitted in outputs.
• CI includes at least one check using SSSOM-native parser/validator.

Related

• Follow-up to current in-branch curie_map-driven cleanup in chromadb_semantic_mapper.py and key consumers.