This is part of meta issue987.

Relies on issue1165.

Referencing issue1165 which aims at removing the non-overlapping assumption from all places outside of landmark factories, i.e., the landmark heuristics and the Cartesian abstraction collection diversification.

Add link to meta issue.

One of the reasons this issue is non-trivial is the existence of the LandmarkGraph::get_landmark(const FactPair &fact) method which requires that every fact is associated with at most one landmark. It's possible to associate each fact with every landmark it occurs in and return a vector of LandmarkNodes in the function instead, but the consequences of such a change are not clear yet.

One of the main reasons we have problems with merging landmark graphs is that 
the code does not support having both A and (A or B) as landmarks, even though A 
or B might have more orderings.

One solution is to get rid of the limitation of having each fact appear in at 
most one landmark.