implement Job.ReclaimablePods for AppWrappers

[dgrove-oss]

In the cases where the appwrapper contains resources that are managed by Kueue and the component implemented ReclaimablePods, we should monitor the workload instances and flow that information through to Kueue.

[dgrove-oss]

This is somewhat related to #74 in that it also relies on fairly detailed understanding of the wrapped GVK's semantics.

[dgrove-oss]

In Kueue 0.7 only Job, JobSet, and Pod (for PodGroups) implement this optional interface. Until the interface is more widely adopted, there is limited value to recognizing this situation and flowing it through the AppWrapper to Kueue.

[tardieu]

While support for Ray jobs seems problematic given #174, we should be able to implement this interface for wrapped PyTorchJobs, in the following sense: if a PyTorchJob status is failed then we can assume no more pods will be created by the Training operator for this job.