kubernetes: kubectl --local is not quite local across all the commands

This situation was discovered when I wanted to remove deprecated functionality in https://github.com/kubernetes/kubernetes/pull/90243. Unfortunately, due to how we are constructing client in almost every kubectl command this removal ended up with the following bug https://github.com/kubernetes/kubernetes/issues/90074.

The reason for that is that in each command’s Complete method we create a client in one of the following ways:

o.Client, err = batchv1client.NewForConfig(clientConfig)
if err != nil {
	return err
}

dynamicClient, err := f.DynamicClient()
if err != nil {
	return err
}

discoveryClient, err := f.ToDiscoveryClient()
if err != nil {
	return err
}

The problem is that with the deprecated code from #90243 removed all of the above calls will fail when creating a client with:

error: Missing or incomplete configuration info.  Please point to an existing, complete config file:

  1. Via the command-line flag --kubeconfig
  2. Via the KUBECONFIG environment variable
  3. In your home directory as ~/.kube/config

which does not happen if we silently default the server to http://localhost:8080 in deprecated getDefaultServer.

There are two possible approaches to this problem:

1. Go through each and every single command and add conditions for --local flag and not to create clients when it is specified.
2. Provide a smart mechanism allowing to inject dependencies (in this particular case, we are talking about clients).

During SIG-CLI call on July 15th, we agreed that solution 1 is a short-term and rather hacky approach to the problem and we would like to see this being approached as described in no. 2.

Any proposals to this should be discussed with @soltysh or during one of sig-cli calls.