Since the remaining action items (LM-cut and merge-and-shrink) are tracked in
separate revisions, I'll mark this one resolved.

OK, goal count heuristic doesn't need action cost support -- counting goals is
counting goals.

This leaves us with three heuristics to consider:

 2. blind: issue192
 9. LM-cut: issue116
 11. merge-and-shrink: issue193

For 3. and 4., see issue186.

Actually, 5b doesn't have proper action cost support yet either. It uses a
bucket-based implementation which is extremely inefficient with large action costs.

For 1. (additive), 5a. (FF) and 10. (max), see issue174.

Implement action cost support for all heuristics. This is a meta-issue. Details
are in separate issues for the different heuristics.

These are our heuristics:
 1. additive
 2. blind
 3. causal graph
 4. context-enhanced additive
 5a. FF
 5b. FF heuristic as part of LAMA-FF synergy
 6. goal count
 7. h^m
 8a. landmark count
 8b. landmark count heuristic as part of LAMA-FF synergy
 9. LM-cut
 10. max
 11. merge-and-shrink
 12. selective max

Of these heuristics, the following fully support arbitrary action costs already
(Erez, please correct me if something is wrong/missing):

 5b. FF heuristic as part of LAMA-FF synergy
 7. h^m
 8a. landmark count
 8b. landmark count heuristic as part of LAMA-FF synergy
 12. selective max