Supporting Sensemaking of Large Language Model Outputs at Scale