Augmenting Human-Annotated Training Data with Large Language Model Generation and Distillation in Open-Response Assessment