Allow bisecting the place where the seed affects program behaviour negatively

[oli-obk]

#1635 made me wonder if we could do something similar to optimization fuel, but for the randomness. Basically if you have one seed that makes your program work and one that doesn't, we give you a fuel parameter that allows you to set the index at which the random number generators get switched, so you can basically bisect to the point at which randomness affects your program in problematic ways (including a backtrace at that switch site)

[RalfJung]

This could help but cannot work reliably I think -- when a certain sequence of non-deterministic (not necessarily random) choices leads to a bad outcome, there's no saying that it was the last choice that is actually problematic. It could be that the very first choice already put the program onto a doomed execution, but it takes a while until the error actually surfaces.

[oli-obk]

there's no saying that it was the last choice that is actually problematic

optimization fuel has the same problem, it could be that the optimization actually breaking things user visibly isn't the problematic one, but gets fed something that was misoptimized by an earlier optimization and is only exposed by the optimization that the fuel bisects to. But at least you then know the "earliest" site which made this problematic.

We can also swap the two random number generators to start with the good one and switch to the bad one when the fuel expires. This suffers from the same problem, but it allows us to potentially find the execution step region in which the problem occurs.