Merging empty partitions branch

[gnikit]

Hi, I was wondering you could merge #162 into master? I did a merge locally and it appears to be fine, no conflicts, no test/unittest failures. I think this would be a great feature to have, although I am a bit biased, since my whole PhD has to do with adaptivity and empty partitions are a real pain to deal with.

I also have a question with regards to this implementation, which I was hoping someone could answer. If you load balance and zoltan returns at least one partition that is empty, does that still cause fluidity to abort?

[jrper]

As far as I remember, this was left still needing code review. I suspect that @stephankramer may still have views on the topic.

The intention (and certainly what I originally coded up the change set to do) is that this version prints a warning when load balancing generates an empty partition, but then carries on with the empty process essentially doing nothing, in hopes that this will eventually be fixed on a subsequent load balancing event. Obviously, this is suboptimal (and somewhat antisocial) on large shared systems.

[gnikit]

It would be great if you would consider merging it into master. I am certainly willing to help with any additional work that might be required for this merge to happen.

With regards to Zoltan and load balancing. I think that having an idle process is orders of magnitude better when you compare it to the alternative of your run randomly aborting after an hour on 10k cores (which was my luck...)