Ingest dbt schema.yml descriptions into AI_CONTEXT

Problem

dbt projects contain rich human-authored descriptions for models and columns in schema.yml files, but the AI_CONTEXT pipeline does not ingest them. Pilot analysts using Dataface against dbt-managed warehouses get metadata that ignores the semantic documentation their dbt teams have already written. This is the highest-value description source for dbt-native users and its absence significantly reduces context quality for the pilot's primary audience.

Context

• dbt model and column descriptions are ingested and linked to AI_CONTEXT entities.
• Conflicts with inferred descriptions are resolved via documented precedence rules.
• Pilot agents can retrieve dbt-authored semantics through MCP context outputs.

Possible Solutions

Plan

• Implement dbt schema parser/mapper for AI_CONTEXT entity IDs.
• Handle missing/renamed model mappings with explicit warnings.
• Add tests for merge behavior and precedence with profile-derived metadata.
• Document ingestion assumptions and required dbt project structure.

Implementation Progress

• Implemented dataface/core/inspect/dbt_schema.py (DbtSchemaParser + merge_dbt_descriptions), wired into MCP catalog via _enrich_with_dbt helper
• 19 tests in tests/core/test_dbt_schema.py covering parser, mapping, and MCP integration
• Two cbox review rounds: R1 fixed dead code, swallowed warnings, not-profiled path skip, added cached-path test; R2 fixed dead logging import and spurious single-table warnings
• 77 tests pass (19 new + 58 existing), no regressions
• Docs checklist item deferred — inline docstrings cover parser/warning semantics
• Description priority merge engine landed in PR #500 (consumer of these sources)

Review Feedback

• Two cbox review rounds completed, all blocking issues resolved

• Review verdict: APPROVED

• [x] Review cleared