investigate cluster capacity prediction

[joel-bluedata]

This is a sequel to the work done for issue #562.

Various sources have requested a "dry run" mode of kdcluster spec creation/update, where KD will reject the creation/update if not enough resources are available to satisfy the entire spec. This does not have to be 100% correct, but it would need to be a very good estimation in order to be a usable mode.

Unfortunately the K8s scheduler itself does not support dry-run, so KD would have to re-create the scheduler logic.

One possible approach for this is for KD to include or otherwise use the cluster capacity tool. Need more research into the state of that project and if/how KD could use it.