This driver connects to an already existing virtual machine,
using the provided IP address and SSH user/key/port parameters.

On this machine, libmachine will provision the docker daemon
and minikube will bootstrap the kubernetes cluster (as usual).

It is not supported anyway, and just throws errors.
There is no use to restart or to retry, just give up.

This should never be a problem with "none", though.
That always return running, while generic tests ssh.

All drivers *must* be in "supportedDrivers"

Delete Podman on Darwin (Mac), not available

It was failing to install the "docker" package

Since the provisioning was defaulting to ISO

Since we don't know the remote machine type anyway,
it could be a virtual machine or a physical server.

But we don't add any _extra_ isolation for minikube,
so go with "bare metal machine" machine type for now.

It was only called for existing machines, not for new.

But other commands expect to be able to run "docker".

This makes the machine driver specific to minikube

The "none" driver has redefined stop and start to mean the
kubelet rather than the host itself. Use the same API here.

Killing actual containers remotely remains to be implemented.
Only stop the kubelet service for now, and assuming systemd.

Not using the default libmachine drivers anymore

Setting the config directly instead in minikube

But most of it is identical to the "none" driver

Now that we have forked the machine driver