Retry farmer lookup on worker start up

[croaky]

When a testbot worker process starts up, it pings the farmer.
If the farmer returns a 404, the worker crashes and does not restart.

Perhaps this could instead involve a few retries? Maybe with a little bit of backoff?
The farmer shouldn't be offline for very long, or take long to start up.
Maybe 1 minute is enough time look for the farmer and otherwise assume it can't be found?