Refactor handling of `Expr` validity

Sadly Cmm is not a total language: there are many programs which one can write which do not have a defined result (e.g. signed division overflow). At the moment we handle this in the Arbitrary generator, taking care to excluding undefined programs in the generator and shrinker. However, this is problematic as it is easy to forget handling UB in either the generator or the shrink (hence #8). Moreover, ideally we would want to have a canonical way to determine whether a Cmm program is well-behaved.

I suspect we should rather have a single function to check the definedness of a Cmm Expr, used by both the generator and shrinker.

mentioned in commit 9eb3ada5

I addressed this in 9eb3ada5 although I stopped short of removing the existing undefined behavior guards from the generators for efficiency's sake.

closed