sort-by-variance

[behinger]

I tried two implementations, one by eeglab, one from the fortran code.

• I think somewhere there is a normalisation step I'm missing, because it seems the variance of all components is always around 1 - that makes it difficult to sort.
• I remember loosely, that in eeglab they are normalizing A by their row-wise (or col?) variance / norm or similar. That should result in different source-activity variances. Is something like that happening in fortran-amica as well and we forgot in Amica.jl?