Decide how to handle CRS transforms

[aazuspan]

Need to decide how best to handle CRS transforms when exporting NAIP imagery, to ensure that predicted maps can be aligned with other rasters.

See @grovduck's comment from #11:

I guess my thought would be this cascade:

1. If CRS_TRANSFORM is defined in config.py, use that preferentially (this is probably what I would do to ensure that all predictions lined up with other covariates in further modeling.)
2. (This one might be tricky because the lidar dataset isn't referenced in 03_export.ipynb). Use the crsTransform of the lidar data that was used to fit the model. This ensures the same snap to the lidar data for prediction and comparison.
3. If neither 1. or 2. is present, use the (projected) corner of the BBOX to set the transform.