Evaluation suite and test datasets

Measure agent accuracy, generate test datasets from schema, and capture conversation feedback.