encapsulate train and predict of PipeOps.
Adding active bindings:

• timings
• log
• warnings
• errors
• encapsulate

and private .encapsulate like Learners have (by default, a Learner's encapsulate is propagated to the PipeOp (if relevant) e.g. in case of PipeOpLearnerCV, PipeOpLearner and as_pipeop.Learner, ...).

A Graph is not encapsulated but naturally it's PipeOps are.

GraphLearners are (already) naturally encapsulated because they are Learners. Additionally they gain the timings_pipeops active binding (closes #536).

Illustration of a GraphLearner timing_pipeops:

task = tsk("iris")
gl = GraphLearner$new(po("scale") %>>% lrn("classif.rpart"))
gl$train(task)
gl$timings

  train predict 
   0.21      NA

gl$timings_pipeops

$train
        scale classif.rpart 
        0.036         0.030 

$predict
        scale classif.rpart 
           NA            NA

Could also add an entry for the time of the glue code e.g., here 0.21 - 0.036 - 0.030 = 0.144

#FIXME: need the following from mlr3 exported:
mlr3:::assert_ro_binding
mlr3:::get_private
mlr3:::append_log
mlr3:::get_log_condition