Adds option for parallel weight generation with xESMF

[norlandrhagen]

With ~0.8 release of xESMF, you can now specify parallel=True to the weight generation which will generate weights with Dask instead of MPI.

Thanks for the xESMF knowledge drop @keewis!

[keewis]

to be more precise, before 0.8 xesmf would compute the weights on a single machine (the jupyterhub instance, for example), and if that ran out of memory you'd have to manually save both source and destination grid to netcdf files and call a command line tool on those to do the computation using MPI (possibly on a bigger machine).