Worker process dying unexpectedly should not result in a zero exit code

I'm seeing this in the logs
```
distributed.nanny - INFO - Closing Nanny gracefully at 'tls://xx'. Reason: worker-handle-scheduler-connection-broken
```

but the nanny process is returning a 0 exit code. This very much feels like it should return a non-zero exit code. Otherwise we can't really differentiate between dask workers that are exiting for sad reasons , and those that are just responding to a sigterm or scheduler scale down request.

Best case the nanny should exit with a variety of documented exit codes depending on the bad exit condition. Next best case pick an exit code and use that for every 'bad' exit.