Move arg registration for `bag_tree()`'s `rpart` engine to parsnip from baguette

[hfrick]

This originally surfaced in censored but basically: baguette registers the model args for bag_tree() with the rpart engine.

https://github.com/tidymodels/baguette/blob/237f2e6df0d8fdc24eed430db4abf3a88fe81f4d/R/bag_tree_data.R#L17-L50

censored also registered that model/engine combo, just for a different mode. So the arguments are only tunable for censored regression if we also load baguette, like we've encountered in extratests:

https://github.com/tidymodels/extratests/blob/307268a1aa4974a3a89248abbcf7d61adfac96c9/tests/testthat/test-censored-case-weights.R#L15-L16

So because both extension package register the same engine, just with a different mode, the shared part of that registration (i.e. things that are not mode-specific) should go into parsnip.

---

The bag_tree() rpart engine for censored regression registers no model
arguments, so none of bag_tree()'s main args (cost_complexity,
tree_depth, min_n, class_cost) are tunable. extract_parameter_set_dials()
and tunable() both come back empty, which means a bagged survival tree cannot
be tuned at all. Every other tree-based censored engine surfaces at least its
tree-control main args, so this looks like a gap rather than a deliberate
choice.

Reproducible example

library(censored)
#> Loading required package: parsnip
#> Loading required package: survival

# All of bag_tree()'s main args, set to tune(), for the rpart engine
spec <- bag_tree(
  cost_complexity = tune(),
  tree_depth = tune(),
  min_n = tune()
) |>
  set_engine("rpart") |>
  set_mode("censored regression")

# No tuning parameters are surfaced
hardhat::extract_parameter_set_dials(spec)
#> Collection of 0 parameters for tuning
#> 
#> [1] identifier type       object    
#> <0 rows> (or 0-length row.names)
#> 

# tunable() confirms the rpart engine registers none
generics::tunable(spec)
#> # A tibble: 0 × 5
#> # ℹ 5 variables: name <chr>, call_info <list>, source <chr>, component <chr>,
#> #   component_id <chr>

Expected behaviour

The tree-control main args that ipred::bagging() passes through to the
underlying rpart fit (cost_complexity → cp, tree_depth → maxdepth,
min_n → minsplit) should be registered via set_model_arg() in
R/bag_tree-data.R so they are tunable, mirroring decision_tree() with rpart.

[hfrick]

We have probably not realized that previously because "it works" if you load baguette before censored. That's because the tuning parameters are registered there - but that's really not the solution.

This has bitten us but we hadn't quite put the pieces together: tidymodels/extratests#186

We can either register those arguments in censored, in addition to baguette, but we should probably just register them in parsnip.