[bind10-dev] optimization: use some additional polling in SyncUDPServer

[Francis Dupont]

I implemented the same kind of optimization in the AFTR: in place of
going back to the main loop (select(), etc) after the handling of
a packet some others are tried with a not-blocking read (so if
there is nothing to do the process falls back to the main loop) and
a counter to not monopolize the CPU. The architecture is a bit different:
there is one thread but some side management activity which must be
interleaved with the packet processing. And packets come through one
file descriptor to the interface/device, a detail which greatly
simplifies the multiplexing.

Regards

Francis Dupont <fdupont at isc.org>