[pptp-server] ugh: real congestion control

[David Luyer]

I started working on real congestion control, but after reading the
draft a bit I realised the way things are meant to work is pretty
stupid and we aren't doing much worse on the outbound by simply
flooding the network.

An ack automatically acknowledges all previous packets in the window,
so if the window is large and there is a strong packet flow, chances
are the window will move quick enough that you will always acknowledge
a range of packets before timeout and you'll just flood the network
in question with dropped packets.

Anyway, I'm going to back out most of my changes and then commit the
useful ones; return the same receive window size on our end as the
transmitter sent us (even though it's a complete lie, the whole thing
seems to poorly designed it shouldn't matter and should improve
performance, an alternative would be returing a constant like 64 or
128).  The other is a fix for a bug to do with failed network
writes, which I don't think can happen anyway.

David.