LLM inference operators on Kubernetes: OME, vLLM Production Stack, NVIDIA Dynamo and llm-dMay 18, 2026