After #19, add a separate test environment to test with dask.
We'd have to introduce a check to make sure that nothing triggers eager evaluation and we return a lazy dataset.
(Also, would be nice to have an idea of performace with an input Bathymetry file such as eORCA36)
