Contrary to previous comments, we can calculate `==` for types
in an extremely general fashion. The approach used here is actually
the one mistakenly rejected as impossible. There will be some cases
when the previous version was able to reduce and this one is not,
particularly for types in `*` that are unknown, but known equal.
However, the new behavior is much more uniform. Within the
established framework of equality testing by pattern matching,
it does a better job than the previous version.

Reviewers: goldfire, austin, hvr, bgamari, RyanGlScott

Reviewed By: RyanGlScott

Subscribers: RyanGlScott, rwbarton, thomie

Differential Revision: https://phabricator.haskell.org/D3835