Use Xarray's `apply_ufunc` rather than Dask's `map_blocks`

[tomwhite]

I've been thinking about how we could run (parts of) sgkit on Cubed (#908). One thing that would help is using xarray.map_blocks (or xarray.apply_ufunc) instead of dask.array.map_blocks, since the Xarray version can now dispatch to Dask and Cubed (as well as plain NumPy).

[tomwhite]

From the documentation xarray.apply_ufunc:

Note that due to the overhead xarray.map_blocks() is considerably slower than apply_ufunc.

Also, xarray.map_blocks is marked as experimental and has not been implemented in cubed-xarray, so for those reasons we should prefer xarray.apply_ufunc. This pathway is being explored in #908.