On the assumption that ArduPilot processes mavlink packets
synchronously (or at least in order), after we have run a timesync
roundtrip we can reasonably expect any mavlink command we have sent
prior to the roundtrip to have been processed, and we should be able to
see the results in the mavlink stream.
This highlights the fact that fetching rally points using the mission
item protocol does some when you're talking mavlink1 doesn't work out
well.
# so this looks a bit odd; the other end isn't sending
# mavlink2 so can't fill in the extension here.
It was observed from a log of a failed CI test that the ACK from
clearing the rally items took 6 wallclock seconds to arrive.
We were not waiting for that ACK to arrive before sending the request
for the mission item count, but if it has taken more than six seconds
for the ACK to arrive it is reasonable to assume that MISSION_COUNT
could very well take more than the 10 seconds we allow for it.
If we drain the mav before sending the request for the mission count we
should remove any signficiant timing problem due to a backlog of mavlnk
messages, but the amount of traffic here is problematic.
Also drain in lots of other places where we might be spending way too
long parsing messages.
