Update acknowledger when we get assignments from the consumer group

[nathanmonteleone]

https://simplifi.atlassian.net/browse/INT-11275

This is a bugfix:

• Consumer worker threads only live for a single assignment. Once assignments are revoked, those threads are destroyed...
• BUT, there is another GenServer called the Acknowledger, that stays alive for the duration of the connection. The Acknowledger tracks the latest offset that it's seen for each partition...
• Normally, when worker threads are started they get their beginning offset from the broker, BUT when they are restarted because of an unexpected crash (say, our json writers timing out) they attempt to restore their offset from the Acknowledger.

In the original code, this can cause a rewind if the worker crashes before actually acking any new offsets.

This PR fixes that by updating the Acknowledger's latest offset as soon as we see a new partition assignment from the broker.