vllm v0.32.0

This release introduces significant improvements to the VLLM project, including structured decision-making and generic parsers, low-precision KV cache enhancements, and a new Model Runner V2. These changes aim to increase throughput, reduce latency, and expand model support.