In 1.4.5, session_timeout acts as a lower bound for request_timeout.

This seems excessively strict. If we use a short timeout like 2s, the heartbeat thread could simply retry, there seems to be some logic already in for that.

In the current implementation, I found these blockers:

• check of arguments
• timeout of heartbeat request makes the coordinator dead. We could remove this for heartbeat thread.
• remove timed-out in-flight-requests from appropriate collections on timeout
• allow receiving responses out of order (Correlation IDs do not match [1.4.5] dpkp/kafka-python#1748)

If we clear these, I think we should allow short request_timeouts.

My goal is to allow better recovery with unstable connections (which do occassionally occur in our environment)

@dpkp what do you think?