Text this: Accelerating Inference in Retrieval-Augmented Generation Models for Long-Form Question Answering via Dynamic Token Pruning.