The conversion to int16 will be done after orthorectofication

by commenting out logs

And check most likelly invalid states first ({0,0,0} is more likelly to happen
than nan's)

Avoid float -> double -> float conversions: works on floats all the way down

... to invalid normals, DEM values...

... to nodata, NaNs, "folded" neighbours (when left and right are a the same
position...)

To avoid clashes with otbIteratorHelpers from OTB