Llm-DDisaggregated Inference on Kubernetes: Routing, Scheduling, and Scaling Beyond One GPUAugust 29, 2025