FEAT: Refusal Scorer

[rlundeen2]

Description

• Adds/generalizes a refusal scorer, which is good at predicting whether or not the response is a refusal
• Updates Crescendo to use the refusal scorer
• Generalizes send_chat_target_async and LLM scorer responses for more maintainable retrying logic
• Adds a retry to LLM scoring if it's expecting a float for a value and the float doesn't parse
• Standardizes timestamp parameter on score (sometimes it was date_time but this is more consistent with other tables)

Tests and Documentation

• Tests for refusal added
• Crescendo and scorer notebooks run