LLM Alignment as Retriever Optimization: An Information Retrieval Perspective