Scaling data for the wdmerger problem on Titan on April 15, 2016.

I ran tests with one, two, and three AMR levels (the higher levels refine
around the stars). For all three I ran in the 4 MPI ranks + 4 OMP threads 
per node configuration, plus a few cases of 2 MPI + 8 OMP for the three level 
test. All data is for 10 timesteps per run.

For the highest number of processors on the two-level run I also used 
the space-filling curve distribution mapping rather than my default 
knapsack mapping, and this was about 15% faster.
