Note [Instances in no-evidence implications]

In #15290 we had
  [G] forall p q. Coercible p q => Coercible (m p) (m q))   -- Quantified
  [W] forall <no-ev> a. m (Int, IntStateT m a)
                          ~R#
                        m (Int, StateT Int m a)

The Given is an ordinary quantified constraint; the Wanted is an implication
equality that arises from
  [W] (forall a. t1) ~R# (forall a. t2)

But because the (t1 ~R# t2) is solved "inside a type" (under that forall a)
we can't generate any term evidence.  So we can't actually use that
lovely quantified constraint.  Alas!

This test arranges to ignore the instance-based solution under these
(rare) circumstances.   It's sad, but I  really don't see what else we can do.