CompositePodGroup API

This KEP describes the evolution in the workload-aware scheduling architecture that is necessary to support more complex, hierarchical scheduling requirements of modern high-performance distributed workloads. We focus on the API, framework and the basic building blocks - performance optimizations of the underlying algorithms can come as follow-ups. To achieve this, the KEP builds on the Workload and PodGroup APIs from [KEP-4671] and introduces a new core API called CompositePodGroup. This API allows expressing multi-level topology constraints, gang scheduling and preemption policies for heterogeneous groups of Pods and facilitates extending the Kubernetes scheduler with more policies in the future.