Alternative libaries for computing cell statistics like scikit-learn

We currently have an implementation of cell statistics using Kirsty's code, leveraged from the regional evaluation suite. However, it has been raised there a quite a few libraries that do image processing of binary fields like scikit-learn, which might be more optimal for identifying structures in precipitation fields like cells. These have the added bonus of being externally maintained, optimised, and scikit package is widely used and available.

This issue is being created to capture this potential library and that scoping of both approaches is needed before we spend a lot of resource on implementing cell statistics.

Links #725 and #329