retrieval: remove the cross-encoder reranker (ablation showed no earned keep) (#100)

eval: built-in rerank ablation, judge on paid hy3, answer-latency measurement (#94)

Eval knobs as workflow inputs: rerank on/off and k (#86)

Add the v1 eval dataset: 100 valid + 100 invalid + 20 agentic

Add a retrieval benchmark: recall@k + MRR over the v0 questions