Review remapping approach

[ThibHlln]

At the moment, the remapping is done by using the two underlying cf.Field of the components' SpaceDomain and the regrids method between the source and destination components' fields. This results in recalculating the remapping weights for every single iteration, which is extremely slow and unnecessary because the geometries do not change between timesteps.

I have raised an issue on the cf-python repository (https://github.com/NCAS-CMS/cf-python/issues/222) to see if they can find a way to cache the weights if repeatedly doing the remapping from already used geometries. This appears to be a limitation of what is offered by esmpy. If this cannot be resolved internally to cf-python, we would need to work on an alternative.

Alternatives I can think of include:

• using esmpy directly in cm4twc to dynamically generate the weights file depending on the components' resolutions, and then hold on to this weight file to do the remapping with esmpy rather than cf-python for each time step
• using esmpy (or any other remapping software) to generate the weights file offline (for a limited number of resolutions, those for Hydro-JULES) and read in this file in cm4twc to do the remapping with esmpy for each time step

[ThibHlln]

cf-python can now return a RegridOperator from regrids/regridc so that the expensive remapping weights generation can be done once and re-used in regrids/regridc in place of destination field. This new feature will be available in cf-python==3.10.0.