Evaluate if we should switch from kubectl commands to API calls

[inercia]

We are currenlty running all the kubectl commands remotely: we upload a valid kubeconfig to the remote host and then run kubectl with that file. This makes things easier in some scenarios, like when the API server is not directly reachable from the machine where Terraform is being run.

However, this is also cumbersome and limits the possibilities for the future. For example, for applying some manifest, we must 1) generate a manifest file 2) upload it to the remote host 3) upload the kubeconfig 4) run kubectl apply in the remote machine 5) delete all these files. This could be straightforward if we could just use the kubernetes client-go and access the API server from the local machine.

And we can think on more complex scenarios in the future, like the creation of machine descriptions for Cluster API and so on...

So we should evaluate if it is really worth keeping the remote kubectl commands or just switch to API accesses...