problem with nodata semantics

[AmitAronovitch]

Generated an skeleton issue so I can link to it from a TODO comment in the current workaround implemented in our code.
Will add details next week.

In short:

(a) Semantics of the "nodata" parameter in the constructor of GeoRaster2 are confusing, and the resulting tiff behaves differently for unit8 rasters than for uint16 rasters (the 16bit case the result is a nodata-based raster, while in 8bit it is mask-based).

(b) I believe we should have a way to generate a true nodata-based 8 bit raster, even if it is not the default, so we can have consistent behavior independent of dtype