SENDME failures on iterated flood/download tests
I believe that there are a pair of linked bugs in our current SENDME cell handling: we count begin and end cells towards towards window totals when we are only supposed to count DATA cells, and we send our sendmes one cell too early (or maybe late?).
I think that these two bugs cancel each other out on the first stream on each circuit, but with the second stream, they cause the circuit to fail