Terminate process correctly from reflector thread

[yuvipanda]

Our reflectors run on their own thread, and sys.exit() from
a thread does not actually exit the process - just the thread.
When the reflector has to shut down because it can not talk
to the k8s master anymore for a while, we had just been stopping
the reflector thread and doing nothing else.

This will instead actually cause the hub to stop, and in z2jh
restart - which seems the right thing to do.

Ref 2i2c-org/infrastructure#680,
where we encountered this during a period of time when the
k8s master was unreachable for about 1 minute.

[consideRatio]

Thanks for the bump @yuvipanda! I hesitated with this during an initial glance a while back and forgot about it.

My initial hesitations were

1. Is this really a desired behavior in the logic?
   Answer: yes - the logic has logging statements indicating that this is really meant to shut down the hub itself, but failed to do so.
2. Will this happen when there is a brief failure of the k8s api-server?
   Answer: no - this happens after several retries and failures.

Conclusion: this is a straight forward bugfix.