New RAG evaluation and LLM-as-a-judge capabilities in Amazon Bedrock