WIP: migrate from QuickCheck to Hedgehog #32

I decided to give #23 a try. So far the two main drawbacks I see is the lack of support for generating functions (there is no analogue of Coarbitrary) and having to explicitly universally quantify with property $ do { a <- forAll genA; ... } instead of defining a property with arguments (though I'm like 80% sure this can be solved with some fancy typeclassing and type families a la printf, it just caught me in a very 1AM point of my life)

I'll take a look at https://dl.acm.org/doi/pdf/10.1145/2430532.2364516 for function generation and see how annoying it is without typeclasses, to keep in line with the way hedgehog does things

It is tempting to do some typeclass trickery, but it plays out rather poorly. Property testing is extremely sensitive to the distributions you choose on the inputs, and classes like Arbitrary funnel you down a path where there’s a single blessed distribution per type. This is why I suggested moving to hedgehog in the first place; there’s much finer grained tools for doing per-test case generators.

What I had in mind was purely code shuffling, something that would turn a MonadTest m => a -> b -> ... -> m () into a Gen a -> Gen b -> ... -> Property. That would handle the quantification for you while still giving you complete control over the distributions. I just want to reduce the boilerplate visible in Internal.Utils.VarSet

Cute diffs!

There's an existing function generation library for Hedgehog (https://hackage.haskell.org/package/hedgehog-fn), but it seems to be lacking in instances. As far as I understand there's no reason not to use a typeclass for "Coarbitrary" because it doesn't affect the distribution of the function generation, though I guess it could affect the shrinking behavior.