Fix a main-thread assertion in the QuorumIntersectionChecker

[graydon]

Change #4240 introduced a new call to assertThreadIsMain(); in the QI constructor. This is correct for the top-level QI checker, but we then run that QI checker on a background thread and, if it does enjoy intersection, we call getIntersectionCriticalGroups which then constructs a pile of QI checkers on the background thread.

The assertion only exists to guard access to the global PRNG to get a seed (the seed used here is a very light requirement that just helps perturb the algorithm's set-splitting path). So a simple solution is to not access the global PRNG in the constructor, but to access it in the scope outside and pass-in a seed. It's harmless to use the same seed in all the sub-checkers constructed in getIntersectionCriticalGroups, so I do that here.

cc @SirTyson

[SirTyson]

Thanks for the fix! I'm wondering if we should wrap gRandomEngine in a light weight class that just checks if the current thread is main on access, it seems pretty easy to accidentally call it from a background thread as it currently stands. A refactor for another day though.

[marta-lokhova]

r+ 68b7f8a