Merged.

Following today's meeting: I didn't review Jendrik's changes again thoroughly,
but from my side this can be merged.

I left some comments. I have a few question marks, but it may be better to
ignore them than keeping this up further, as I'm not sure I'll find time to
discuss them this week.

Was more directed at Jendrik. But I agree that this doesn't have to be answered
or dealt with here.

Not sure if the question is directed at Jendrik or me, but generally speaking
the landmark code is in dire need of more general polish. I would be somewhat
reluctant to invest much time in local improvements before having looked at
things more globally; they may turn out to be wasted effort in the end.

So focusing primarily on the memory management here makes some sense to me. But
of course that's always a case-by-case decision.

I left a few comments. Overall, this looks good.

I was wondering, though, if we still need so many places where we use raw
pointers of landmark nodes for hashing and/or comparing landmark nodes (I
assume). Basically, we now control the landmark nodes using unique_ptr and we
store them in a vector instead of a set, but we have still plenty of places
where we access and also store the raw pointers. Do you think that this could be
changed at all? If so, it might be a different issue of course.

Silvan, are you up for a review?

The results are in:

https://ai.dmi.unibas.ch/_tmp_files/seipp/issue899-v1-opt-issue899-base-issue899-v1-compare.html
https://ai.dmi.unibas.ch/_tmp_files/seipp/issue899-v1-sat-issue899-base-issue899-v1-compare.html

The patch makes the landmarks code considerably faster. This leads to solving
four more intances for the lm_hm configuration. Otherwise coverage is not
affected much.

Here is a scatter plot for the total_time needed by BJOLP:
https://ai.dmi.unibas.ch/_tmp_files/seipp/issue899-v1-opt-issue899-base-issue899-v1-total_time-seq-opt-bjolp.png

I'll have a look once the experiments are done.

I made a pull request at 
https://bitbucket.org/jendrikseipp/downward/pull-requests/124 .