Optional use of listcredentials for uploading CNF images to Artifact Store

[sunnycarter]

Work item

This changes the way image and helm chart upload is done for CNF. The default behaviour is the same as before, i.e. CLI user context permissions are used for helm push and image copy (import). This will fail if the user does not have subscription-wide permissions (documented in params.py and various docstrings).

It adds two optional features:

1. source_local_docker_image specified in the config file. Only works for a single image. Means you don't need a remote ACR if you're just doing a quickstart or noddy CNF
2. --no-subscription-permissions CLI flag. Changes behaviour to use manifest credentials for helm push and docker image copy (now via docker pull, docker tag and docker push). Requires docker to be installed. Slower but doesn't require subscription-wide permissions in order to access Artifact Store.

Additional changes:

• Using az acr import instead of python ContainerRegistryManagementClient, because we shell out all the time anyway, and that client was a bit of a pain to use, and required knowledge of resource group names. Without it we can just ask the user to provide the ACR login server name.
• Fixes to various exceptions to use ones that don't show the whole stack trace to screen
• Small tidy up of artifact_manifest vs artifact. We don't use the concept of an artifact client for helm push or image upload (we didn't before). I haven't fixed this but I've had to add manifest credentials to artifacts so they are available.
• We can't check that the source ACR exists before we start now, as we don't know the RG. A small downside but we will get a decent error I think.

Testing:

• Single helm chart and image, source image, source ACR import, source ACR docker push pull
• Multiple helm charts and images, same
• Misspelled source_registry gives decent output
• Bad permissions (had to fake as don't have the right subscription access)

[jamiedparsons]

Looks good! Thanks for improving a whole bunch of output text - that sort of thing makes a real difference.

Do you think that there is any risk of regressing Windows support here? Worth a quick test?