parsnip::fit_xy() should accept a 1 column data frame for y

So you don't have to force it to be a vector. This is especially relevant when juicing a recipe and subsetting to `all_outcomes()`, where a data frame with 1 column is returned for univariate models.