The relaxed problem does not need these. It would only throw off gringo later.

Since those variables are going to be assumptions, we do not want them
to be derived.

This is a bit of a hack, but it's the quickest way to get aggregates to
parse. We don't need to inspect them, since they remain unchanged in
relaxation.

There's no benchmark here

Using Graph Coloring and Combined Configuration from ASP Competition
2015, slightly altered.

This requires some more testing, but works on the basic example in choice.lp

This still doesn't parse all LARS atoms. An atom token should also
contain any window descriptions etc.