Revisit 0c_Data_Preprocessing.ipynb to prepare rasters on common grid with common extent/dimensions

[dshean]

https://github.com/uw-cryo/DeepDEM/blob/main/notebooks/0c_Data_Preprocessing.ipynb

A lot of this can be done with vrt, no need to read and write out lots of additional files with slightly different extents.

Right now most of your inputs are pre-processed using identical CRS and resolution by @ShashankBice's scripts. If this is not the case, then I'm concerned about "off-by-one" dimensions in the output from this:

width = np.round((window_read_bounds[2] - window_read_bounds[0])/reference_x_resolution).astype(int)
height = np.round((window_read_bounds[3] - window_read_bounds[1])/reference_y_resolution).astype(int)

'compress': 'deflate', is not necessarily the best option for rasters with continuous float values. Use default of LZW. Should be faster and offer better compression.

The use of resampling=Resampling.nearest can introduce interpolation artifacts - this should always be bilinear or bicubic.