Add engine specific predictor encodings

[juliasilge]

This PR makes workflows aware of the new engine specific predictor encodings implemented in tidymodels/parsnip#319 and discussed in tidymodels/parsnip#290 (and elsewhere).

The main idea is that instead of adding a blueprint to a preprocessor in a new_action_ function, the blueprint argument is passed on and evaluated later, within .fit_pre() in the new `update_model_encoding() function.

• If the blueprint has been set by the user, no updating takes place. 🚫
• For the formula preprocessor, we add the default blueprint and then look up the model encoding and update it. 💃
• For the recipe preprocessor, we only add the default blueprint (no relevant encodings). ✅

The tests use a ranger model because it is more clear what the "right" values should be, but no ranger function is called within .fit_pre() (only parsnip functions).

In one test, we have:

expect_equal(ncol(result$pre$mold$predictors), 6)

If we fix 🤞 the issue around one-hot encoding vs. indicator values, this should go to 5.

[DavisVaughan]

Thanks!

[github-actions]

This pull request has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.