Avoids ending up with huge log directories.

plus some param alphabetic reordering

Align the way motor output is computed with the model used in the mixer generation script.
previous:
`out = (roll_gain*roll + pitch_gain*pitch + yaw_gain*yaw + thrust)*out_gain`
new:
`out = roll_gain*roll + pitch_gain*pitch + yaw_gain*yaw + thrust_gain*thrust`
which is the standard matrix*vector multiplication.

This commit has 0 effect on symmetric platforms (i.e. most) as thrust_gain==1 on these platforms.
On asymmetric platform --where some thrust_gain are lower than 1-- the previous formulation resulted in coupling between thrust and other axes.

fixes #8628

The geometry was previously quad_deadcat in which front motors are closer to CG and thus more loaded in hover.
quad_wide is the same geometry as quad_deadcat except the CG is centered so all motors are loaded equally.
Flight logs on IRIS with deadcat mixer showed that
- all motors are equally loaded during hover (actuator_outputs 0 to 3 have similar values)
- a negative pitch offset is building up soon after takeoff (visible in actuator_controls)

- Missing get_sync() in verify function
- Allow some time to reboot before closing the serial port
  (closing the serial port drives TX low)

This is a temporary workaround for equal expressions in chip/sam_xdmac.c
and can be reverted once fixed properly in NuttX. Needed for the GCC 7
upgrade in #8551.

 - revert after #8689 is resolved

without this param set doesn't work -> adsp side never receives changes

If a log with multiple (dis)arming events was replayed, multiple logs were
created. This changes the replay to only create a single log.