Timelimit for ensembles

[mb706]

• input pipeop that starts timing but does nothing
• wrapper learners that are connected to that by-reference
• somehow use timeout by encapsulatoin? or at least refuse to start learning once timelimit is reached
• then, idk, train fallback learner maybe? or return some kind of NULL result
• may need a kind of error object in the graph that gets forwarded and collected, maybe similar to NOPs