Move initialisation of SerialManager and GCS up to AP_Vehicle

[peterbarker]

Functional changes here:

• AP_Stats is initialised after we initialise the serial manager and the GCS console. AP_Stats doesn't do anything that warrants being initialised before SerialManager, and if things to wrong in stats init it would be good to get diagnostic output (if any)
• Sub sets the delay callback much earlier. A good thing as it is more consistent with the other vehicles and may give GCS feedback while calibrating.
• we set the ins log bitmask bit rather later in Plane. Considering we initialise logging much than this code, it doesn't matter

[khancyr]

I love it !
This need a small override fix on rover

[khancyr]

AP_Stats could be moved too, can't it ?

[peterbarker]

@khancyr AP_Stats is protected by #ifdefs. We only get to use the HAL_ optional compilation things in libraries. Whether AP_Vehicle counts as a library... nobody knows ;-)

[peterbarker]

The Rover thing is more than just adding the override. That method is doing some seriously funky stuff to do with outputs, and will require a chunk of inspection before moving the call up in the initialisation order. It also should be removed from Rover's init_ardupilot method.

So one thing I notice is that set_control_channels in Rover calls functions on the motors library before the motors library is init()'d. Fun stuff.

[peterbarker]

I've looked through the Rover change and think moving the set_control_channels should be OK. The calls made in there are configuring ranges and types on input and outputs, which should affect things only in a positive manner.

... still needs to be run on some vehicles.

[peterbarker]

I've now put this on a Rover and it didn't cause any fires.