Exponential backoff retries logic never kicks in if redpanda is down

[vncorpacc]

We upgraded from version 2.3.5 to a version 2.4.6 and noticed this regression.

On version 2.3.5 we relied on exponential backoff retries (configured to Infinite) for our environment restart operations in case redpanda-console starts up before redpanda. Those worked properly as coded in kafka.NewService.

Apparently this commit: d06eaa3
breaks this behaviour by attempting a "admin client connectivity" check in redpanda.NewService without any retries. Failure to establish it is fatal.