Agent Failure, Trust Repair, and Fluency in Human-AI Teams