Leveraging Passage Embeddings for Efficient Listwise Reranking with Large Language Models