Leveraging Approximate Caching for Faster Retrieval-Augmented Generation