* Defer netDb searches for newly referenced peers until we actually want
them
* Ignore netDb references to peers on our shitlist
* Set the timeout for end to end client messages to the max delay after
finding the leaseSet, so we don't have as many expired messages floating
around.
* Add a floor to the streaming lib window size
* When we need to send a streaming lib ACK, try to retransmit one of the
unacked packets instead (with updated ACK/NACK fields, of course). The
bandwidth cost of an unnecessary retransmission should be minor as
compared to both an ACK packet (rounded up to 1KB in the tunnels) and
the probability of a necessary retransmission.
* Adjust the streaming lib cwin algorithm to allow growth after a full
cwin messages if the rtt is trending downwards. If it is not, use the
existing algorithm.
* Increased the maximum rto size in the streaming lib.
* Load balancing bugfix on end to end messages to distribute across
tunnels more evenly.