Improve performance of Predict GAM/GLM tools by replacing ASCII --> Raster mechanism

[jjr8]

Implemented in r740, released in MGET 0.8a24.

I did not end up using rgdal to write the output rasters. The problem with that is GDAL cannot write all of the formats supported by ArcGIS (including, for example, ArcInfo Binary Grid format, which has been the standard in ArcGIS for years). Also, rgdal does not currently expose all of GDAL’s options through R which makes it very difficult to create anything that is fully ArcGIS-compatible (for example, GDAL requires you to set a special option in order to write signed 8-bit integers, which will commonly be needed for classification tree models, but rgdal does not provide a mechanism for setting that option).

Instead, I changed the R code to write temporary binary files which the MGET Python code then converts to the destination raster using an efficient mechanism. If the destination format is one that the Python version of GDAL can write in a way that is fully compatible with ArcGIS, then GDAL is used and it is extremely fast. The Python version of GDAL does expose all of GDAL’s options, so the main formats that cannot go through that route are only ArcInfo Binary Grid and ArcGIS geodatabase formats. For those formats, the ArcGIS Copy Raster tool is used, which is still much faster than the existing implementation.