AMQP Drain of a link encounters unexpected response delay

Copying from https://issues.apache.org/jira/browse/ARTEMIS-1283 upstream by tabish121

When an AMQP attempts to drain a link the broker can take a inordinate amount of time to respond that the link is drained, especially if there's no messages to send. This happens because the action that attempts to force delivery is signaled of completion via a callback and the update to proton is made but no flush of the connection state is done.

As a side not the drained call to the link is done outside the connection lock which is unsafe.