Make it possible to use Python functions outside of ESMValCore as preprocessor functions

[bouweandela]

It could be convenient if users could define their own preprocessor functions and use those from the recipe instead of being limited by what is available in ESMValCore. For example:

preprocessors:
  # See https://docs.esmvaltool.org/projects/esmvalcore/en/latest/recipe/preprocessor.html
  # for a description of the preprocessor functions.

  annual_mean_amsterdam:
    extract_location:
      location: Amsterdam
      scheme: linear
    annual_statistics:
      operator: mean
    esmvaltool.diag_scripts.example.some_function:
      example_argument: 10
    multi_model_statistics:
      statistics:
        - mean
      span: overlap
    convert_units:
      units: degrees_C

where esmvaltool.diag_scripts.example.some_function is a Python function in the module esmvaltool.diag_scripts.example with the function signature

def some_function(cube: iris.cube.Cube, example_argument: int) -> iris.cube.Cube:

This would make using the preprocessing capabilities of the recipe more powerful because recipe authors can include their own Python functions that may not need to be of general interest.

@ESMValGroup/technical-lead-development-team Opinions anyone?

[valeriupredoi]

must say: bonus points for the extra flexibility this may bring in! But...

publishing a new recipe will be a bit tricky in this instance since the processed data will be unique to the custom and external function used by the author, we ought be careful with that; there's another aspect to it - Dask integration may be problematic if the user has a certain configuration for it to best suit their custom function; lastly, how are we gonna be able to test such an external function for long term maintenance?