Allow raster_to_numpy_array to return arrays from overviews

[phargogh]

In the NCI water quality pipeline, it's convenient to be able to read overviews for some rasters, especially for use in a matplotlib graphic which will be downsampled anyways, rather than the full-resolution array. We can do this with GDAL, but it might be nice to have some of this boilerplate in a nicer function.

        factor = 2  # determined by another process
        pygeoprocessing.build_overviews(
            raster_path, resample_method='near', internal=False,
            overwrite=True, levels=[factor])

        raster = gdal.Open(raster_path)
        source_band = raster.GetRasterBand(1)
        if factor == 1:
            mapp = source_band.ReadAsArray()
        else:
            # We built overviews with factor 4, but these are available at the
            # 0th overview layer.
            overview_band = source_band.GetOverview(0)
            mapp = overview_band.ReadAsArray()

I'm thinking that it might be nice to be able to request `pygeoprocessing.raster_to_numpy_array(path, overview=2) and then have it determine the correct overview based on the resolution.