Laws

[armanbilge]

Thoughts:

• The laws for Random are probably best tested with something like TestU01?
• Probably should introduce a type e.g. GenSpawnRandom that extends both Spawn and Random and has laws asserting the statistical independence of the spawned fibers. In fact, this should probably extend down the entire Spawn hiearchy, and when users want both Random and Spawn/Concurrent/Temporal/etc. they should really ask for e.g. ConcurrentRandom assuming they want this guarantee of statistical independence (i.e., almost always).
• There is some trickiness for MonadError through MonadCancel due to loss of state in error/cancellation channels. A naïve StateT-based implementation of Random would have independence issues here. But I think the only way to observe this issue would be via some side-effect that can only be implemented with Concurrent or Sync. So maybe also a SyncRandom type is needed, with corresponding laws to enforce statistical independence in these cases as well *grumble*.
• Finally, what to do about the distribution families. I wonder if their laws can be defined in terms of the laws for Random i.e. it should be possible to define Random in terms of any distribution by transforming via the CDF.