My theory is that it's ok to send the `noidle` synchronously but the
subsequent handling of the idle response should be on the command queue.
My hope is that this fixes issues with less than perfect network
connections (and also very occasionally seen on good networks).