add sparse argument to step_dummy()

[EmilHvitfeldt]

This PR adds the creation of sparse dummy variables in step_dummy() via the use of the sparse argument.

[EmilHvitfeldt]

it is set to default to FALSE because it should only be used if you need it. BUT in workflows we plan to detect if the model accepts sparsity, then toggle sparsity in steps that can do it.

Example helper function below. Currently we don't have the infrastructure yet to determine if the sparse columns produced by step_dummy() will make it unharmed through. If step_normalize() was placed after step_dummy() in this example, the sparse dummies would be ruined. But we might be able to do so in the future. For that reason i think the helper should live in {recipes} instead of {workflows}. Any objections @simonpcouch?

library(recipes)
library(modeldata)

recipes_all_sparse <- function(x) {
  for (i in seq_along(x$steps)) {
    if (!is.null(x$steps[[i]]$sparse)) {
      x$steps[[i]]$sparse <- TRUE
    }
  }
  x
}

rec_spec <- recipe(Sale_Price ~ ., data = ames) |>
  step_unknown(all_nominal_predictors()) |>
  step_impute_mean(all_numeric_predictors()) |>
  step_normalize(all_numeric_predictors()) |>
  step_dummy(all_nominal_predictors())

rec_spec |>
  prep() |>
  bake(NULL) |>
  lobstr::obj_size()
#> 7.46 MB

rec_spec |>
  recipes_all_sparse() |>
  prep() |>
  bake(NULL) |>
  lobstr::obj_size()
#> 1.71 MB

^{Created on 2024-11-12 with reprex v2.1.0}

[simonpcouch]

Super clean!

No objections for that helper living in recipes rather than workflows.

Good move on scoping this PR as just implementing step_dummy(sparse) and holding off on tackling the "smart" toggling the argument in the backend. Your thoughts about what that interface might look like sound reasonable!