Originally, the code used isc_nm_pauseread(). During adaptation to the latest Network Manager changes, it has been replaced with isc_nm_read_stop(). However, there is no need in that as currently in order to continue reading, we need to explicitly call isc_nm_read() on the underlying TCP socket. So, this call is superfluous at the moment, if not harmful, considering that other parts of the code appear to be properly adapted.
The isc_nm_read_stop() is known for causing troubles (assertion failures) in the Stream DNS branch when paired with new read timer control functionality there, in the development branch it might be responsible for some intermittent failures too.