Minimum 2 observations per category in all permutations

Hi guys, 

If you have a binary variable with a low count on one of the categories and you use a lot of permutations, it will eventually happen that one permutation does not have 2 or more counts on that category and glmnet throws an error.

It could be nice to catch this and exclude these permutations, but provide a warning.

Best,
Jonas
