Proposed title of this feature request

Annotation to allow cluster autoscaler to ignore rescheduling simulation for unmanaged pods

What is the nature and description of the request?

Request for support/implementation of upstream autoscaler PR https://github.com/kubernetes/autoscaler/pull/9134.

This would provide support for the annotation cluster-autoscaler.kubernetes.io/skip-reschedule: "true".

As stated in the upstream PR, this annotation would allow for pods with this annotation to be evictable during scale down and not simulated for rescheduling.  This would allow for unmananged pods that don't need destination simulation to be terminated when the node is removed.

Why does the customer need this? (List the business requirements here)

The new Mountpoint for S3 CSI driver v2 from AWS puts unmanaged pods on nodes, and the cluster autoscaler fails to remove these pods during scale-down operations because of failing rescheduling simulations.

There is an existing issue in the AWSLABS project (https://github.com/awslabs/mountpoint-s3-csi-driver/issues/675).  However, the customer would also like this ability to explicitly skip rescheduling for selected unmanaged pods as an alternative solution for this particular Mountpoint for S3 CSI driver issue as well as similar future issues.

List any affected packages or components.