Send all packets with the same DSCP field

[remip2]

irtt --dscp=N sends the first packet with DSCP 0 which can cause issues with stateful network devices on the path which register dscp=0 for the udp pseudo-connection. This fix sets the same DSCP field on the first and following packets.

[heistp]

Thanks for the pull request. I'll think about taking in this change, but one thing to note is that in the case where the server doesn't support DSCP (or disables it with --no-dscp), the open packet will contain a non-zero DSCP, and each additional packet will contain a 0 DSCP, so this will also confuse any stateful middleboxes which are classifying traffic in this way.

I'm not sure that middleboxes should be using the DSCP value of only the first packet as a class identifier. However, I see that at a minimum, pfSense is doing so (see note under Caveats section: https://docs.netgate.com/pfsense/en/latest/trafficshaper/diffserv.html). The problem with this is that the DSCP value can change at any time as the flow goes through different routes, which may treat packets differently, and may even wipe the dscp field entirely.

So at a minimum, we'd need to stop the client fallback in the case where the server doesn't support DSCP, but I'm still thinking through what the right overall solution is here...