feat: minimum relevance threshold for retrieval #10
Description
Context
From #6 — the same facts appear in retrieval results regardless of query content.
Problem
Retrieval currently returns top-k results with no minimum relevance cutoff. With a small knowledge base, every query returns the same entries because there aren't enough entries to differentiate, and low-scoring results are never filtered out.
Proposal
Add a minimum relevance threshold so that only results above a configurable score are returned. Consider also adding a diversity mechanism to avoid redundant results dominating the output.