Problem: messages dropped due to HWM race

[bluca]

Using ZeroMQ in a system with throughput of several million messages per second I was observing occasional drops of small number of messages, even when HWM is set to a value sufficient to accommodate all possible bursts in the flow.

The problem seems be caused by a race between process_activate_write() and check_hwm() on peers_msgs_read variable in zmq::pipe_t.

[bluca]

Fixed by #1649

This fix was backported to 4.1, so using this issue for the NEWS.