helm: retry on connection refused

[3u13r]

Proposed change(s)

• Let the helm installer retry on "connection refused" errors

Problem:
constellation init fails with:

Your Constellation master secret was successfully written to ./constellation-mastersecret.json
Initializing cluster   
Cluster initialization failed. This error is not recoverable.
Terminate your cluster and try again.
Error: init call: rpc error: code = Internal desc = initializing cluster: installing pod network: helm install: Kubernetes cluster unreachable: Get "https://X.X.X.X:6443/version": dial tcp X.X.X.X:6443: connect: connection refused

Detailed description:

What is interesting that after kubeadm init we have an explicit kubewaiter, which waits until it can list all K8s namespaces using the kubeclient and further kubeclient requests such as annotating the current node. But the execution of the first helm command fails.

The kubeclient automatically retries on almost all error cases. The helm client just on timeout errors (which the error we see is not, it is a "connection refused" error).
During all those requests mentioned above all targets (i.e. nodes) are unhealthy. This means that the AWS LB "fails open", routing to all targets.
This becomes most apparent on AWS since the switch from unhealthy to healthy takes at least 20 seconds. On GCP and Azure it should take 2-5.

Checklist

• Add labels (e.g., for changelog category)
• Link to Milestone

[netlify]

✅ Deploy Preview for constellation-docs canceled.

Name	Link
🔨 Latest commit	c9da251
🔍 Latest deploy log	https://app.netlify.com/sites/constellation-docs/deploys/63f5d720af82ac000737ece9

[derpsteb]

🟢 AWS, 1.25, 2+3

[katexochen]

🟢 AWS, 1.25, 2+3

Doesn't seem to be related to this PR, but https://github.com/edgelesssys/constellation/actions/runs/4234338625/jobs/7356490507#step:9:3468

[katexochen]

lgtm