Always acknowledge from worker to agent to spot lost packets.

Currently the requests from agent to worker are broadcast as a single udp packet which can be > frame size so they get split up.  If one of the packets is lost this can lead to long delays until it is retried.

The workers already send acknowledgements in some situations (e.g. I'm already working on it), so it should be easy to extend to all situations, and to have a shorter retry if the message does not appear to have been received.

(Does the receiving code also logic to cope with udp messages where part of the message has been lost, or will it just be thrown away?)