feat: cBioPortal + OMOP ingest pipeline and Databricks bridge

[deanban]

Problem

Evaluating source-semantic-hardening required a realistic healthcare corpus. Existing connectors only read generic Databricks metadata — we had no way to land cBioPortal study data (SV, CNA, gene panel matrices, resources) into Databricks or to model OMOP CDM targets for semantic mapping.

Proposed solution

• Add cBioPortal parsers for structural variants, copy number alterations, gene-panel-matrix, and resource files.
• Add OMOP CDM model coverage so target-model mapping has a first-class schema to map to.
• Build a Databricks bridge with DuckDB staging for iterating on parsed study data locally before landing it.

Delivered in commits 137066f and 18f13f4 on branch dean/feat/source-semantic-hardening.

Alternatives considered

• Mock cBioPortal fixtures — rejected; wouldn't exercise the Databricks connector path or real grain edge cases.
• Skip OMOP and only evaluate on raw cBioPortal — rejected; target-model mapping is the forcing function for Stage A grain and Stage B semantic typing.

Closed by #63.

[deanban]

Closed by #63