An open handbook on LLM inference at scale (GPU internals, KV cache, batching, vLLM/SGLang/TensorRT-LLM) [P]