Optimize interaction between circuitmux and circuitpadding
In legacy/trac#29204 (moved), we realized that we needed to inspect the circuit queue to prevent the circuit padding system from queuing too much padding. For that ticket, we're going to do the simple fix -- we don't schedule padding if the circuit's cell queue already has a window full of cells on it. In that case, we just lie to circpad and tell the machine we sent padding even though we did not.
However, we should determine if we want to have a more complicated interaction, such as delaying our padding until the queue drains. We should also examine creating alternate machine events for when cells are dequeued from circuitmux vs when they are sent locally. This has pluses and minuses for machine interaction.