Move time series aggregation to an external module

[Stefan Pfenninger]

Problem description

To reduce complexity of Calliope's core code, we only want a hook for time series aggregation and resampling, rather than actually doing it ourselves.

The external module could be:

• Our own current code moved out of the Calliope core
• tsam

TODO:

• Remove all complex clustering algorithms from core (inc. masking).
• Move time resampling to model.resample_time.
• Make it possible to cluster the timeseries using a user-defined set of cluster IDs (the functionality already exists, we just need to move the definition to model.cluster_time.
• Keep a config to switch enable inter-cluster storage when using clustering (e.g. model.include_inter_cluster_storage, default is True).
• Update docs to tell people to prepare cluster IDs themselves using e.g. tsam.
• Make hardcoded sum/mean of data on resampling explicit for every input parameter.
• Move hardcoded sum/mean of data on resampling (calliope/time/funcs.py:294 ea89a66) to a model_data variable attribute (ideally, this would be encoded in the typedconfig rules).
• Document justification for sum/mean of input parameters on resampling.

[Bryn Pickering]

In the context of #452, we could now have config.init.time_resample alongside config.init.time_subset.

We could also move these two configuration items to config.build and allow a user to resample/slice data only when they build the optimisation problem?

As I see it, advantages:

• Quicker initialisation of the model as we aren't doing any timeseries manipulation
• ability to test different extents of resampling / time subsetting on-the-fly
• Can save the initialised model to file and load it later to do different timeseries operations

Disadvantages:

• larger model when input data is long, although time_resample would have no impact here as currently when we resample we keep a copy of the original timeseries in-memory anyway.
• odd output timeseries / possible clashes in output. If resampling, one would get gaps between timesteps. If subsetting, one would get gaps either side of the subset.

[Stefan Pfenninger]

We have decided not to provide clustering code for now, and leave it up to users to do clustering as per their requirements. As of 0.7, it's possible to supply user-defined clustering: e.g. config.init.time_cluster: cluster_days.csv