Union types are parsed and stored as a raw string (e.g., "type": "'primary' | 'secondary'"). This is human-readable but not machine-usable / LLM-optimized.
Proposed State: We should explicitly parse these strings into a clean array (e.g., "enumValues": ["primary", "secondary"]).
Why it's critical: This makes the data structured and reliable for an LLM.
Action Item: Modify cli/tsDocGen.js to handle this.