Tune algorithms for conflux
I need to run a bunch of shadow sims to tune the conflux algorithms for latency, throughput, and mem usage, and then update the spec for the new algs.
As part of this, we also need to improve the throughput for simulated Hong Kong endpoints. For some reason, they did not improve as much as Germany did. I suspect the problem is that only switching once per cwnd may be handicapping this throughput for higher latency links. We probably need to allow it to switch more often.
When this is done, we also need a changelog entry: #40780 (closed)