Inside vLLM: Following One Request from the API to GPU Execution

Follow one vLLM V1 request through scheduling, paged KV-cache management, GPU execution, and output delivery.