Automated testing of TimeSteward callers

[elidupree]

There are various ways that TimeSteward callers can misbehave, which we should find ways to audit for.

• Using nondeterministic code in predictors or events.
• Using field types that have lossy serialization or don't have an exact match between Eq and serialization equality.
• Using unsafe_now() improperly.
• Using (column, event, predictor) types that are not included in Basics::IncludedTypes.
• Using nonrandom data in (column, event, predictor) ids.

[elidupree]

One of the important techniques for testing determinism is to run 2 instances of the same simulation on different devices, and continuously verify that they remain synchronized – even with network delay affecting the fiat events issued by each device. This requires a full TimeSteward implementation, and also some additional features we haven't settled on yet.

[elidupree]

simply_synchronized is now working and successfully caught me using HashSet inappropriately. There is much more work to be done in this area – including making the panic messages more detailed and understandable – but I consider this sufficient for MVP.

[elidupree]

With the API changes, the old testing is no longer applicable, and there are many new tests that need to be done.

My idea is to make an auditing TimeSteward – a separate TimeSteward implementor that does costly operations to detect as many possible incorrect behaviors as possible. I'm thinking it would also be able to take a replay – a record of fiat events plus a record of what order all events were run – and rerun the simulation in that order, even if the original simulation was run by a real-time TimeSteward instead of the auditing one.