Evaluating Quality of Answers for Retrieval-Augmented Generation: A Strong LLM Is All You Need