This fixes #796.
It also fixes the @bboston7's example on #883, but not the original
example in #883.   The issue there is that `asdf` function generates
a constraint only involving the module parameter.

Normally we don't try to solve constraint that don't mention a schema's
parameters, because they can always be solved "later", and hopefully with
more things instantiated.

The weird thing in this case is that the schema adds *local" constraint to
the module parameters, but we end up ignoring these.

When checking a signature we try to solve sub-goals that mention any
of the variables mentioned in the signature's assumptions.  We do this
because these assumptions are local, and if the assumption is not solved
in the current context, we may not be able to solve it later.

This is OK if the goal does not contain free variables.  If it does,
things are a bit flaky as we may reject some programs and accept others
depending on in what order things happened to be instnatiated