Change Applicative GenT to use zipping

[jacobstanley]

This changes the Applicative instance for GenT so that it zips the tree rather than binding it. This means that a generator like (,) <$> genA <*> genB will be able to alternate between shrinking genA and genB rather than doing one first and then the other.

An example shrink tree using a zipping approach vs just using (<*>) = ap:

I had it like this originally and changed it because it broke the law <*> = ap, but I'm not sure this is that important as the Monad instance for GenT is already unlawful.

Haxl also uses an unlawful Applicative instance for basically the same reason. This was the motivation for ApplicativeDo which you would be able to use with Hedgehog after this patch to achieve more optimal shrinking.

A nice article explaining another option and the tradeoffs: https://elvishjerricco.github.io/2016/09/17/abstracting-async-concurrently.html

I also renamed Tree to TreeT, so I could have Tree = TreeT Identity as I was working with TreeT Identity a lot during testing and it got a bit tedious.

[jacobstanley]

hmm can't use CPP #ifdef with language extensions, weird

[jacobstanley]

It's wild that the scala commits are showing up here, pretty cool

[jacobstanley]

I decided requiring transformers >= 0.5 was a lesser evil than dropping 7.10 support.

7.10 has been a real pain recently though, these last few PRs always seem to have some issue.

It is pretty old now! Version 7.10.3 (released 8th December 2015)

[moodmosaic]

I guess we should consider dropping GHC 7.* when GHC 9.* comes out, which might not be so far away given the increased GHC release cycles.