Add runtimeClassName support for pod specs in the K8s plugin

[alexpilotti]

New feature

The K8s executor does not support specifying the runtime class for pods. This is needed when a K8s cluster provides multiple container runtime configurations.

Use case

A very common use case consists in accessing a K8s cluster with a separate runtime configurations for containers that access NVIDIA GPUs, where running the pod with the default runtime would cause an error as it is not configured for this purpose.

Suggested implementation

The suggested implementation consists in adding runtimeClassName to the pod directive settings.
For example:

process {
    executor = 'k8s'

    accelerator {
        limit = 2
        type = "nvidia.com/gpu"
    }

    pod = [ runtimeClassName: 'nvidia' ]
}