Merged.

The change is simple enough not to require performance evaluation, I think. (I
may have said otherwise before, but it's always a bit different if you're
talking about a change you don't know about versus a concrete diff.) So I'll
merge this now; thanks everyone!

Regarding performance evaluation in general, I think we should set up some
"checkpoint" experiments based on the IPC 2011 configurations. Maybe I can get
one of the hiwis to set up the basic infrastructure for this.

The changes look good to me and were overdue (this was still a fragment from the
old bucket-based vector implementation). 

Malte, I once suggested to make this change but you said, I may not without
running experiments to measure the impact on the performance. Are you still
worried about that?

Regarding the "last_evaluated_value": For this, we probably should implement the
"evaluation contexts" that also encapsulate the related information in the
search implementations.

Done, but not yet merged. Gabi, can you do a quick code review?
http://codereview.appspot.com/4641065

Jordan, if you want you can pull the changes from the issue246 branch.

Gabi, do you have some time next week to do a general review/design meeting
about the open list and evaluator code? I think there are some overdue cleanups
there. I'm particularly worried about the "last_evaluated_value" etc. fields,
which seem quite fragile and at least need documentation. (It'd probably be
better to get rid of them entirely.)

Jordan made me aware of some things in the scalar open list implementation that
we might want to clean up.