Rework the data scaling algorithm for the "deadzone" mode so that the
deadzone is subtracted from the input rather than from the output. This
makes the whole output range usable rather than making the output jump
from the center value to the minimum deadzone range.
This changes the calibration data structure as well so now all values
refer to the input data, which is more coherent.
Signed-off-by: Fabio Baltieri <fabiobaltieri@google.com>
