Erez's initial experiments were very promising. I've merged the code now.

Erez, can you add documentation to the wiki?

Apart from that, we're done with this for now. (I think that conjunctive LM
support can be added a later time in a separate issue, if desired.)

I've done an initial implementation, which doesn't support conjunctive landmarks.

I will run some tests overnight to see if it gains anything.