Deadlock can occur when multiple concurrent disconnects trigger a reconnect

[cpforbes]

When a an error occurs and the client is unexpectedly disconnected from the broker it will automatically reconnect.
If this occurs when responding to multiple concurrent requests each thread will trigger a reconnect via the on_disconnect MQTT callback. Thus multiple on_connect and on_disconnect callback threads will be running concurrently.

It can occur that an on_disconnect callback thread is holding the ServiceManager Rlock and is attempting to acquire the _connected_lock in DxlClient (via the connected method) while at the same time an on_connect callback thread is attempting to acquire the ServiceManager lock (via ServiceManager.on_connect called from _on_connect_run)

This only occurs when many concurrent requests all trigger the broker to disconnect the client as it does when a oversized message is received. 6-10 concurrent requests will usually trigger the deadlock.

This is related to and/or the cause of #55