Write a TimeSteward that takes advantage of parallelism about time-steward

[elidupree]

We may need to add some API parallelism-support features for predictors and events.

When an event or predictor gets invalidated while it is still running, it would be nice for it to save time by exiting early.

Moreover, it would probably be more efficient to discard invalidated fields than to preserve them for predictors/events that are in process. The natural way for the accessors to handle this is to have get() return None, which would mean that you can never safely unwrap() the result even if you know that it will always be present in the legitimate simulation. We could provide a "unwrap or return" macro. (try_get!() seems like a good name for it)