Surv => Regr pipeline using Buckley-James imputation

[bblodfon]

This is the (old) code that did the task transformation (the bj$ynew is the imputed times, see R package bujar):

.bj = function(backend, status, time) {
      require_namespaces("bujar")

      x = data.frame(backend)[, colnames(backend) %nin% c(time, status), drop = FALSE]
      x = model.matrix(~., x)[, -1L]

      bj = invoke(bujar::bujar,
        y = backend[[time]],
        cens = backend[[status]],
        x = x,
        tuning = FALSE,
        vimpint = FALSE,
        .args = self$param_set$get_values(tags = "bj")
      )
      backend[[time]] = bj$ynew
      return(backend)
    },

Regression learner would predict the survival times, so we can map that back to the response survival prediction type and set crank = -response (which the PipeOpPredRegrSurv had done incorrectly)