This may cause unexpected behavior when the outer control loop emits a zero (and eventually perhaps negative) reference in response to a disturbance. This is why arming commands need to be decoupled from the setpoints.
We could introduce an option that will make the drive refuse to start while the setpoint is zero; this is quite easy to do, but it may not aid in building safe systems.
I think both solutions will require decoupling hover and tractor ESCs into separate topics, because if you pursue the first option you will still use a separate topic for controlling the arming of a subset of ESCs. Also consider that not publishing commands to the hover ESCs during cruise flight conserves network bandwidth and processing power (and a tiny bit of energy, not that it matters here). So I would go with the second option as it is more general.