read_image() should allow users to select a subset of the image they want to read, like load_image() already does
currently load_image() supports (in probably a buggy way) rewriting things like coordinate dask arrays to numpy arrays, attr dask arrays to numpy arrays, sub-xds dask arrays to numpy arrays ... as well as data variable dask arrays to numpy arrays. Some of these conversions are questionable and are done on the current xds (not a copy). For now, we really only need top level data variable dask arrays converted to numpy arrays, as these are what our node_tasks operate on. So, just convert those for now
read_image() should allow users to select a subset of the image they want to read, like load_image() already does
currently load_image() supports (in probably a buggy way) rewriting things like coordinate dask arrays to numpy arrays, attr dask arrays to numpy arrays, sub-xds dask arrays to numpy arrays ... as well as data variable dask arrays to numpy arrays. Some of these conversions are questionable and are done on the current xds (not a copy). For now, we really only need top level data variable dask arrays converted to numpy arrays, as these are what our node_tasks operate on. So, just convert those for now