instead of single I/O request per step, the daemon now retries
all addresses in the selection with 300ms timeout between tries.
there are len(list) + len(list)/2 tries

the idea is to reduce latency when UDP request doesn't punch through,
or some NSs are overwhelmed/faulty