Consider support for dask and lazy loading

[markpayneatwork]

Is your feature request related to a problem? Please describe. KAPy is built upon xarray and xarray supports dask for parallel reading, processing and writing of data. It's a great tool and potentiallycan give substantial speed improvements

Describe the solution you'd like Provide support for dask in KAPy, either by default or via a switch

Describe alternatives you've considered There are some issues to consider

• snakemake already takes care of parallelisation to some degree by running multiple jobs in parallel. It's not immediately obvious that dask could give further speedups for most applications
• dask can potentially run into problems with chunking that lead to massive performance deterioration, but can be very hard to debug. This may exclude it from being a suitable tool for the KAPy software stack

[markpayneatwork]

This commit https://github.com/Klimaatlas/KAPy/commit/d6380d6c95b1fa17ddce5461f2c3f427d8590464 restructures the import functionality to avoid xarray automatically using dask. At this point, KAPy is therefore an in-memory processing tool. We may need to fix this in the future.

[markpayneatwork]

dask and lazy loading are also very closely coupled - for some datasets, it may be advantageous not to write out the intermediate file, but instead just return the xarray object. How this works best with cdo sellonlatbox subsetting is unclear, but the two things need to be thought together at the same time