Close issue #13: half the size of models when using cvglmnet classifiers by *not* storing the call within the models.

Fix issue #14: used which for unmapped features as well and then a set operation to make sure that f.idx will not become empty if there is no overlap.

fix issue #103: update p.adjust methods, adjust xlab for signficance plot if mult corr method is none.

updates for confounder check: check that the metadata does not have too many levels, re-written some error messages, bugfix for base confounder plots, add variance plots.