Tooling: tighten CDS/HANA answer path to avoid whole-file .cds reads

[jung-thomas]

Problem

When CDS lookups via cds-mcp miss, the agent falls back to reading multi-thousand-line .cds files to find an entity, field, or endpoint — pulling far more into context than the answer requires.

Proposal

Tighten the CDS/HANA "answer" path so the agent never needs whole-file reads:

• Ensure cds-mcp reliably returns just the entity/field/service/endpoint for this project's models (audit misses).
• Consider a thin project-scoped index of entities → fields → @path endpoints → source location, returning only the matched slice.
• Same idea for HANA schema lookups via hana-cli (return the column/table, not a dump).

Impact

Smaller, incremental win vs #1–#3, but removes a recurring large-file-read pattern.