Optimizing Rare Word Accuracy in Direct Speech Translation with a Retrieval-and-Demonstration Approach