apply PipeOpTaskPreproc to data.frame / data.table

[mb706]

• Add another function, e.g. $operate(), which uses the train_df() functions internally and accepts data.frame? (Should this set a state? Maybe $operate(df, train = TRUE)?)
• Just have $train() and $predict() accept df?
• Do the same with Graph?
• have the %>>% operator work on df and do the necessary conversions, so that df %>>% po("removeconstants") works?

Things to consider:

• character cols in df often "mean" factor-cols in Tasks.