Results changing after some updates

[ddundo]

Previously I noticed that results from goalie demos would change over time and now saw it on the bubble shear case here. When the bubble shear demo was merged, the classical adaptation results were 31.76% relative L2 error with 2302.3 average number of vertices. Rerunning the demo now gives 30.73% and 2314.3. There haven't been any other commits to animate since then. Uniform resolution results stayed the same.

Obviously minor, but I'm curious what is the likely cause. I guess it's something on the mmg side? If so, I think it might be a good idea for us to track major changes happening in mmg and document them somewhere as well (in the wiki perhaps), since we rely on it.