In FF, EHC did not do global duplicate detection.  (Issue originally identified 
by Maximilian and Joerg)
The current implementation in FD does. 

The attached patch adds a flag for choosing the correct behavior.
We might also want to eliminate cycles from the plan.

I am closing this; the intention to make our EHC implementation more "standard" is good, but I think there are other, more substantial differences that would be more important to address (e.g. the lazy search nature of the implementation), and for this I think it would be better to keep the logic simple.

Also, with our current state space implementation, at the level of the search space we must necessarily do global duplicate elimination anyway, so the form that "does not" do it actually does some kind of hybrid, which I think is hard to explain, too.