When a peer restarts and a backoff is applied locally, it needs to be
reset once the peer is available again. Otherwise the transport does not
reconnect until the entire backoff elapses thus marking it unavailable
for downstream consumers like partition balancer.

Adds node_status_reconnect_max_backoff_ms cluster configuration
and defaults to 15s.

Wrap it in a waiter. With debug builds the backend can potentially
take some additional time to reach the desired state, especially
after invalidating all the transports after resetting a backoff.