Feature request: support for channels spanning chunks

Hey Jeremy, in #238 you mentioned:

> However, Neuroglancer currently only supports that if the dimension is unchunked, i.e. the chunk size is equal to the full extent of the dimension; for your example dataset, the extent of the "c" dimension is 4 but the chunk size is 1, so it can't be displayed.
>
> Potentially this limitation could be removed, though.

Support for channels spanning chunks would be useful for supporting [ngff/ome-zarr](https://ngff.openmicroscopy.org/latest/). While the specification does not require channels be split, most of the current tooling does ([bioformats2raw](https://github.com/glencoesoftware/bioformats2raw), [ome-zarr](https://github.com/ome/ome-zarr-py)).