Enable distributed cube computing

[forman]

Is your feature request related to a problem? Please describe.

xcube serve supports datasets that are computed on-the-fly, e.g. applying temporal aggregation. This requires high data throughput and CPU resources, therefore requests based on computed datasets often respond too slow and then time out. Such computations may be much faster if computation is distributed on a dedicated cluster.

xcube gen may be parallelized so that individual input files (usually spatial time slices) are transformed in a distributed way and then results are combined chronologically into the desired cube.

The same may apply to other xcube commands that perform heavy computing on chunks of data.

Describe the solution you'd like

Add option to xcube CLI commands that configure how the command is executed in a distributed manner. The option is TBD.

Describe alternatives you've considered

None, besides using larger machines.

Additional context

As we use xarray and xarray uses dask, the solution should be based on Dask Distributed.

[forman]

Started this so we can start using Dask.distributed for first tests.

[forman]

Evaluate Dask-distributed for multi-user