In our e2e tests, we see a lot of "etcd-leader changed" errors
while deploying non-essential helm charts.
If this transient error occurs, helm gets into a broken state
where it cannot uninstall cleanly and thus any retry attempts fail.
By not waiting for the installation of helm charts to succeed,
we can avoid making most of the kubernetes API calls while
control-plane nodes are joining.
This makes "constellation apply" faster and more resilient.
9d164de18b