K8s driver improvements

[carolynvs]

I am working on a Porter Operator that can run CNAB bundles using K8s CRDs. As part of that I had to make some changes to the Kubernetes driver to get real (not sample) bundles to work.

• Mount a persistent volume to share data: injected files and outputs, between the driver and the bundle job.
• Apply custom labels in the kubernetes driver Allow the caller to specify additional labels to apply to resources created by the driver.
• Fix setting CLEANUP_JOBS I had the if statement inverted... I've added a test to make sure it (and other settings) are correct.
• Remove requiredCompletions from k8s driver The only value that ever makes sense for requiredCompletions is 1 because we aren't running parallel pods for the bundle job.
• Improve defaulting for kubernetes driver
  • When LimitCPU or LimitMemory is 0, do not set a limit. We don't set a request size, and requesting 0 is not the same as not having a requested resource quantity. This can lead to increased pod evictions.
  • Clarify that LimitCPU and LimitMemory are also used for the requested resources.
  • When ActiveDeadlineSeconds is 0, do not set a deadline.
  • Split reading the configuration in SetConfig and attempting to connect to the cluster, so that we can unit test the settings logic.
  • Move validation for KUBE_NAMESPACE into SetConfig so that any errors can be reported immediately.
  • Add comments for all of the driver settings

With these changes I am able to run @squillace 's kubeflow bundle! 🎉

apiVersion: porter.sh/v1
kind: Installation
metadata:
  name: kubeflow
spec:
  reference: "ghcr.io/squillace/aks-kubeflow-msi:v0.1.7"
  action: "install"
  credentialSets:
    - aks

Closes #73

[carolynvs]

requiredCompletions was not exported, so there wasn't any way to set it to anything other than 1. So we had this configurability in place but there was no way to use it (in addition to it not making any sense to change either).