Process collections of model development runs for each model_id.

For some model development experiments, we need to consider simulations in "collections", for which we have two or more simulations for each `model_id` (i.e. each science configuration).

For example, we examine atmospheric climate feedbacks by using the difference between a pair of experiments: amip and amip-p4k. Or some analysis may require an ensemble of amip or historical runs for each model.

Can CDDS standardise data in MASS ensemble data classes? The UKCI people have done this for PPEs, but did this require a customised build/configuration of CDDS?

No files, tests, or entry points are named. Start by tracing how CDDS currently handles MASS ensemble data classes and compare that with the UKCI PPE setup, including whether it required a customized build or configuration. Done means the required collection behavior and any configuration changes are clearly specified.

Written by the indexing model from the issue text.