As it happens, this is a pretty major change, because I misunderstood
how the protocol actually works. Rather than having a single core
command channel and then a series of offshoots, SOCKSv5 does a separate
handshake for each individual command, and then uses the command stream
as a data stream. So ... whoops. So now the `SOCKSv5Server` sits on a
listener, instead, and farms each of the connections out to a task.
The prior implementation involved some (moderately awkward) layered
`match` expressions; this version skips those in favor of `?`. The cost
is a little less detail about (for example) when serialization errors
happen. On the bright side, it gains us some huge improvements in code
clarity. We might be able to get back the tracing later, with other
library support.
In addition, we know have a couple tests: one to make sure we choose the
authentication method appropriately, and a very basic handshake test for
when we're not actually negotiation a username and password.
This adjusts the way that TestingStream is implemented to allow for
two, separate buffers for each of the two directions. In the prior
implementation, if you called `write` and then `read`, you would
`read` the data you just wrote. Which is not what you want; you want
to block until you get data back from the other side.
I believe these were introduced previously to solve a problem that we're
no longer dealing with; specifically, if I remember correctly, we
introduced these to deal with how we were going to implement a trait.
However, they don't appear to be necessary any more, so we're going to
get rid of them, so we won't need to deal with them any longer.
This induces an added `Send` in the Network trait and its
implementations, but provides us the ability to use standard functions
with obvious extensions. So that's nice. I've also added some additional
testing to sanity check the conversions.
