ability to specify runtime as argument to the .deploy method?

[RafalSkolasinski]

It may be cleaner to specify runtime as argument to the .deploy method so we can avoid modification of the pipeline instance state.

This way it will be more transparent what remote we deploy to - and also possible to deploy to two remotes at the same time.

For example:

# Talking to docker runtime
docker_runtime = classifier.deploy(runtime=DockerRuntime(...))
docker_runtime.wait_ready()
docker_runtime.remote(payload=np.array([[1, 2, 3, 4]]))

# Talking to k8s runtime
k8s_runtime = classifier.deploy(runtime=KubernetesRuntime(...))
k8s_runtime.wait_ready()
k8s_runtime.remote(payload=np.array([[1, 2, 3, 4]]))

One present problem is that information about upload is included in the artefact itself in the .save and then .upload calls.

[ukclivecox]

No needed anymore