Discover skolems in the hypothesis, not just goal

[isovector]

The tactics plugin was only discovering skolem types present in the hole, rather than anywhere in the hypothesis. #541 gives the following repro:

foo :: Monad m => (a -> m b) -> (b -> m c) -> (a -> m c)
foo f g a = _

here, the hole has type _ :: m c, which means tactics doesn't realize a and b are skolems, and instead treats them as unifiable variables. Thus the insane solution of f a.

Fixes #541

[jneira]

That was a really quick fix!