Is Your Paper Being Reviewed by an LLM? A New Benchmark Dataset and Approach for Detecting AI Text in Peer Review