Provence: efficient and robust context pruning for retrieval-augmented generation