Add kind signatures to tyvars with kind not 'Type'

[harpocrates]

This means that the kinds do not get completely swallowed up in
signatures where they matter, for instance in

 foo :: forall proxy (a :: ()). proxy a

It also means that all higher-kinded type variables now have
explicit kinds.

/cc @sjakobi

---

I don't think this is what we ultimately want, but we definitely want at least some of these kind annotations. I'm open to implementing any ideas people may have for constraining the number of annotations we show.

The FunArgs test is what initially motivated me to make these changes:

• On ghc-head:
  

• On wip/hi-haddock:
  

• On this branch:
  

[harpocrates]

I've come up with what I think is a decent heuristic for hiding some kind signatures and foralls. In particular, we drop kind signatures for type variables when

• the type variable has at least one use site in the type where it is fully applied
• the fully applied use site has kind Type
• at the fully applied use site, the the (kind) arguments of the kind of the type variable match exactly the kind of the arguments to the type variable

For example, consider the type variable m in mapM :: (Traversable t, Monad m) => (a -> m b) -> t a -> m (t b). Here is the logic around not needing a kind annotation for m:

• take the m b use site
• m b :: Type
• m :: Type -> _ and b :: Type

Finally, if all of the type variables in a type end up not needing kind signatures (and the order in which the type variables are mentioned matches the order of the explicit foralls), we drop the whole forall ....

Working on wip/hi-haddock, it turns out that these heuristics cause almost all kind signatures on functions to disappear. (Those on data declarations and classes remain - but I like that!)

Screenshots

For the FunArgs test case, we get:

[sjakobi]

Great stuff! :)