Add better support for logical outcomes

[gabrielodom]

I ran some bootstrap samples, and the Type I error is preserved even if we supply a binary outcome as our "continuous phenotype". This is because we invert the test, and put the outcome on the RHS of the model (the methylation has always been on the LHS). However, we coded this assuming that the phenotype would always be continuous. So now, I have to make sure to transform any binary phenotypes to numeric in R first.

Adding support for other categorical outcomes is also mathematically valid, but would be more challenging to implement.