Explore loading speed and subsetting for global orca domain

[b-barton]

Loading global data and taking a subset for computing heat content on 1 year of data takes varying time according to the method:

• with COAsT it took over 1h15 (it actually didn't finish because of a bug),
• with Xarray and the preprocessing partial function it took 12m55,
• with Dask it took 6m47.

Does this bug in (1) still exist?
Investigate feasibility of implementing speed ups in COAsT.

AC:

• Create ticket for bug
  

• Propose ticket for speeding up COAsT users (ideally a solution using COAsT.gridded() but worse case e.g. just ADDITIONAL documentation about how to use Dask.