VerticalPodAutoscaler does not reconcile VerticalPodAutoscalerCheckpoint properly, causing wrong podMinCPUMillicores being calculated for pods

Issue

• We found that VPA would recommend the CPU value to be set to 50m for a specific pod. But when changing the container name in the deployment, suddenly, two VerticalPodAutoscalerCheckpoint are being reported and the minimum CPU is set to 25m instead of 50m, which is not correct.
• podMinCPUMillicores value is set to 50% on the respective deployment even though it has only one container and therefore should have 100% of podMinCPUMillicores set.
• VPA aggregates not pruned when container names change causing stale results
• VerticalPodAutoscaler does create VerticalPodAutoscalerCheckpoint resources for each Container in a pod. Yet it does not reconcile the object in case the container name in the pod is changing, causing podMinCPUMillicores to be applied on the base of two containers in one pod instead of just one.