Drop unused factor levels in categorical variables

[aalfons]

Unused factor levels in categorical (independent or control) variables should be dropped. Argument drop.unused.levels in function model.frame() controls this and should be set to TRUE.

[aalfons]

Likely this needs to be fixed in the default method of fit_mediation(), where non-numeric x variables and covariates are converted using model.matrix().