Comparator is unnecessarily resource intensive

[mdbooth]

The comparator gets the schema for an object's GroupVersion on every reconcile:
https://github.com/openshift-cloud-team/boxcutter/blob/c0d5c889bc4c068e315cda223c5d5bf1d3c62fee/machinery/comparator.go#L200

It then has to parse it before using it in the comparator. Most of these will be from a small number of GroupVersions. For example, in my test set of 84 CAPI objects there are only 5 GroupVersions. There is no caching afaict, so each GV's schema is re-fetched and re-parsed many times per reconcile.

Even worse, the interaction of server-side defaulting with atomic structs and lists means that the comparator frequently returns Modified when the object hasn't actually been modified, meaning that the redundant SSA it is trying to avoid happens anyway.

The performance impact of this doesn't appear to be too awful in practice, but it's probably not good for the environment. Would be nice to improve this.