Rearchitect/refactor runtime options to provide consistent mappings

[axsaucedo]

Currently if the following code is run, the model created would still be running on Docker instead of kubernetes (main difference being the RuntimeOptions vs SeldonCoreOptions)

from tempo.serve.metadata import RuntimeOptions, KubernetesOptions, InsightsOptions
from tempo.serve.constants import DefaultInsightsK8sEndpoint
from tempo.seldon.k8s import SeldonCoreOptions
from tempo import deploy

kubernetes_options = RuntimeOptions(
        k8s_options=KubernetesOptions(
            namespace="seldon",
            authSecretName="minio-secret"
        ),
        insights_options=InsightsOptions(
            worker_endpoint=DefaultInsightsK8sEndpoint
        )
    )

remote_model = deploy(pipeline, kubernetes_options)

The following code would be the one that would run correctly

from tempo.serve.metadata import RuntimeOptions, KubernetesOptions, InsightsOptions
from tempo.serve.constants import DefaultInsightsK8sEndpoint
from tempo.seldon.k8s import SeldonCoreOptions
from tempo import deploy

kubernetes_options = SeldonCoreOptions(
        k8s_options=KubernetesOptions(
            namespace="seldon",
            authSecretName="minio-secret"
        ),
        insights_options=InsightsOptions(
            worker_endpoint=DefaultInsightsK8sEndpoint
        )
    )

remote_model = deploy(pipeline, kubernetes_options)

We should re-explore the different options in runtime and how the combination of these would map into their respective deployments. Initially there are other things to address such as there not being a SeldonCoreOptions in seldon/docker.py, but generally it would be a good oportunity to review the available dataclasses.

This coudl also be a good opportunity to reduce the different ways in which runtimes can be used, and simplifying the code paths.