[wish] more stable calibration calculation

[tias]

instead of averaging the click-coordinates and then computing the min/max values, we could compute the min/max values using the clicks separately, and average the min/max values itself.

instead of min = calib( (x1 + x2)/2 )
do min = ( calib(x1) + calib(x2) )/2

[tias]

We could, but it would not matter (much)
The reason that calibrations can be unstable is that it is possible to use a calibration that registered a misclick.
When issue #3 "detection of rogue click" is implemented it will be impossible to do a bad calibration.