How to handle tuning with `glmnet`

To use `tune()` in `{tidymodels}`, you have to supply potential values of `lambda`. The way `cv.glmnet()` works is to first call `glmnet()` on the whole data, then do the splits using the sequence from the full fit.

So the proper way to do this with `{tidymodels}` is simply to mimic that: fit all training data, then pass the path along.

We should incorporate this into the vignette in #105 and #131.