Fix GKE

[mclasmeier]

I finally wanted to fix deployments to GKE when using containerized roxie.

I figured out it fails due to two distinct problems:

1. The installation of the gcloud SDK was broken due to an architecture mismatch. This is fixed by moving the gcloud install step to the runtime stage.
2. roxie kept failing during initialization, I don't completely understand why, but my testing has shown that when doing something like

podman run --rm -v $KUBECONFIG:/kubeconfig:U -v $HOME/.config/gcloud:/.config/gcloud:U localhost/roxie:latest env

it kept failing -- the initial attempt to use the GKE auth plugin failed because it was lacking write permissions to some files (several!) in ~/.config/gcloud, even though I instructed ownership translation using the :U mount flag. Hypothesis: there is likely to be some kind of race caused by the container starting up even though the file permissions in ~/.config/gcloud are not really effective immediately during container startup. I have verified that adding a dumb delay mitigates this. So, this PR also contains a simply retry mechanism for the initialization, which seems to do the trick. See:

❮ podman run --rm -v $KUBECONFIG:/kubeconfig:U -v $HOME/.config/gcloud:/.config/gcloud:U localhost/roxie:latest env 
00:00 Running containerized.
00:00 Using kubeconfig /kubeconfig
00:00 Attempt 1/3 failed: failed to retrieve API resources: exit status 1, retrying...
00:00 Using kubeconfig /kubeconfig
Roxie Environment Information:
==============================
Kube config:          /kubeconfig
Running in Container: true
Current Context:      gke_acs-team-temp-dev_us-central1-a_mc-04-08-unhappy-danger-soci
Cluster Type:         GKE