Save posterior moments even when moments are constant

In covariance_mc_analysis.m and the like we test whether the moments are constant and, if yes, we store NaN. I don't see the logic of this. All moments are still well-defined (although identical). I would propose to get rid of this check and always store the computed moments.