NVIDIA’s NVFP4 KV Cache Revolutionizes Inference Efficiency