Uniform treatment of partial state in node evaluation

[geisserf]

Issue imported from old repository.

Currently, a partial state can have zero domain assignments for a given variable. In this case, we assume that the whole domain is covered. However, this leads to a non-intuitive special case in evaluation, where we have to check if the covered domain is empty and then treat it as a fully covered domain. It would be better (and lead to cleaner code) if we assume that a partial state has at least one value for every variable. For min/max evaluation, we can just create a partial state which covers all domains for all variables and give it to the partial evaluation.

[digitalw]

agree