TaskSurv and beyond: add support for other survival tasks and censoring types

[RaphaelS1]

Low priority until we come upon a working use-case.
Infrastructure in place with TaskSurv but need to find learners that can actually handle this in R.

[bblodfon]

After some initial personal experimenting and a discussion with @adibender and the rest of the survival group:

• We need a different Task type per survival problem: recurrent events, time-dependent covariates, multi-state, competing risks, etc.
• The censoring type follows from the problem (which survival::Surv() types can be used and what methods can be applied). This means mstate and I think the counting types should be removed from TaskSurv. Right, left and interval censoring should remain (with some further checks for the interval, ie one subject id per row in the data).
• Methods such as $times() and $status() should be expanded/refined to filter for individual risks when type = "mstate" for example in a TaskCompRisks or TaskMultiState
• Time-dependent covariates, how to handle?