Function extract_response_pairs 

pub fn extract_response_pairs<'a>(
    baseline: &'a [Record],
    candidate: &'a [Record],
) -> Vec<(&'a Record, &'a Record)>

Extract (baseline_response, candidate_response) pairs by pairing the i-th chat_response in baseline with the i-th in candidate.

If the counts differ (e.g. candidate had backend errors), truncate to the shorter of the two. Callers that need divergence-on-count should consult the replay_summary record directly.