Remove hardcoded `VectorDistribution`s from survival learners

[bblodfon]

Two examples (there may be more):

• https://github.com/mlr-org/mlr3extralearners/blob/main/R/learner_partykit_surv_ctree.R
• https://github.com/mlr-org/mlr3extralearners/blob/main/R/learner_partykit_surv_cforest.R

[RaphaelS1]

@jemus42 this will fix the cforest bug you found

[bblodfon]

Note: the only other hard-coded VectorDistributions are in surv.parametric (https://github.com/mlr-org/mlr3extralearners/issues/41) and flexsurv (https://github.com/mlr-org/mlr3extralearners/issues/318), both to be solved in other issues.

[bblodfon]

@jemus42 it is done, you shouldn't run into any type of problems with this learner (and it should be fast to calculate metric on predictions from those now)

[jemus42]

Awesomesauce, thanks! 🙏🏻

The thing is, I've realized one of our issues is the serialization of objects with environments, so the other two learners (parametric and flexsurv) also still affect my experiment as long as their prediction object includes a $distr slot that's not just a data.table 😬

[bblodfon]

flexsurv I will fix soon - for the parametric, we have to ask Raphael, cause it really returns VectorDistributions and not matrix-like object inside the $data$distr slot