proposal/bugfix: Remove checks for gdal resampling algorithm values

[emlys]

Fixes #387

This was pretty simple so I went ahead with it, but happy to consider alternatives too!

We were relying on a gdal bug to build a list of human-readable names for resampling algorithms. In gdal 3.8.5, this no longer works. We could implement a different way to get the names, but I propose that we remove the list entirely.

The lists in question (_GDAL_WARP_ALGORITHMS and _GDAL_WARP_ALGOS_FOR_HUMAN_EYES) are used in warp_raster and build_overviews. They're only used to check that the provided resampling algorithm is a valid option, and raise a helpful error message if not. I think we can do away with that because GDAL raises its own (slightly less helpful) error message if given an invalid algorithm name:

gdal.WarpRaster raises a TypeError: in method 'wrapper_GDALWarpDestName', argument 4 of type 'GDALWarpAppOptions *' and prints out ERROR 5: Unknown resampling method: <resampling method>. With gdal.UseExceptions(), you get a nicer RuntimeError: Unknown resampling method: <resampling method>.

gdal.BuildOverviews raises a RuntimeError: Building overviews failed or was interrupted for <raster path>. and prints out ERROR 1: GDALGetResampleFunction: Unsupported resampling method "<resampling_method>". With gdal.UseExceptions(), you get RuntimeError: GDALGetResampleFunction: Unsupported resampling method "<resampling method>".

These error messages aren't quite as nice as what pygeoprocessing had, but they're adequate to identify the issue. Since we pass the resample method argument directly to the underlying gdal function, I think it makes sense to let GDAL handle it.

A couple of behavior changes:

• The gdal.WarpRaster resample algorithm argument can be a gdal.GRA_* constant as well as a string name. By removing the pygeoprocessing check, we can accept gdal.GRA_* constants too.
• gdal.WarpRaster and gdal.BuildOverviews use the nearest neighbor algorithm if the resampling argument starts with 'near'. So 'near', 'nearest', 'nearest_neighbor', and 'near_foo' all work.

[phargogh]

Yeah, this makes sense as a solution. In GDAL 4.0 the switch to exceptions by default will make it more readable anyways, and this solution is a more general one since, as you point out, allows GDAL to decide how to interpret a warp mode and whether it is invalid. Thanks @emlys !

Just waiting to confirm that the tests pass before merging.