Watching node in remote clusters should use a cache

[JoelSpeed]

Note: this is only applicable once #2250 is merged

Watching of nodes in remote clusters was introduced as part of the Machine Health Checking work in #2250.

Currently it uses individual node informers from client-go to subscribe to events for nodes within the remote clusters, however, it could use the newly merged controller-runtime source type for kindWithCache which is designed for this purpose (kubernetes-sigs/controller-runtime#794) which would make the implementation more controller-runtime native and a bit simpler too.

Once kubernetes-sigs/controller-runtime#794 is available in a release, the dependency on controller-runtime should be updated and this should be implemented

/assign @JoelSpeed

[JoelSpeed]

I've already had a go at this, which I can add as a WIP PR once #2250 is merged.

I need to test it on a real cluster though