This one is kind of interesting. We should detect at install time if we're running in an arcpy environment. If we are, skip adding GDAL as a dependency (this is just a list comprehension in setup.py, not sure about other package specs). Then we can run raster merging and zonal stats without additional installations and with less code that runs faster. But we still want to support the current pipeline since it is more cross-platform and doesn't require a license. We'll want to see if we can test the arcpy branch and we'll want to compare the outputs with current methods to make sure we get roughly comparable numbers (pandas has a method for asserting two data frames are roughly equal down to a certain decimal precision. We'll use that)