Nogood-Learning for complete semantics (3-valued)

[ellmau]

Is your feature request related to a problem? Please describe.
Right now the nogood-based learning algorithm is only capable of reasoning for two-valued models (i.e. 2-val models and stable models).
Complete (and preferred) semantics cannot be solved due to the 3-valued nature of them.

Describe the solution you'd like
An enhanced version of nogoods, to work with 3-values.

Describe alternatives you've considered
The current implementation of the naive approach is the only alternative right now.

Additional context
None so far, correctness and robustness need to be shown formally too.