Kafka client does not handle consumer group coordinator changes

[graphcareful]

The observation of this bug came from this CI failure where it was observed that pandaproxy returned 500 due to a sleep aborted exception not being handled.

Upon further observation with @BenPope it was discovered that the test was actually hanging before the 500 was returned due to the client being stuck in a retry loop due to a group coordinator change.

To fix #5638 we must have the client make a new find_coordinator request when not_coordinator is observed in consumer.cc