Choose implementation backend for hybrid retrieval follow-up to #44

[voidrot]

Summary

Make the implementation decision for the first hybrid retrieval backend, using #44 as the planning anchor and current storage constraints as the decision boundary.

Why this work is needed

The repo already has a hybrid-retrieval plan, but implementation should not start until the backend choice is explicit. The current direction should compare a Postgres-first path against Chroma only as a concrete implementation decision, not a default assumption.

Scope

• Evaluate a Postgres-first implementation path for lexical and vector retrieval.
• Evaluate whether Chroma adds enough value to justify an extra operational surface.
• Clarify lexical search expectations, including whether exact BM25 is required.
• Produce an implementation recommendation for the first shipped path.

Out of scope

• Full retrieval implementation.
• Migrating documents into Postgres.
• Replacing markdown as the source of truth.

Acceptance criteria

• The first implementation backend is explicitly chosen.
• The decision documents why markdown remains the source of truth.
• Follow-on implementation issues can proceed without reopening the backend debate.

Dependencies

• Depends on Epic: Hybrid retrieval implementation over reconstructable indexes #60.
• Follow-up to Plan hybrid retrieval over markdown with BM25 and vector search #44.