[MLton] Minor front-end extension

[Stephen Weeks]

> I was thinking it could be very useful for us to have an option
> which rejects toplevel structure definitions without signature
> constraints.

This feels like a very strong restriction.  I guess the motivation is
to prevent accidentally exposing identifiers that should be hidden?
Would you require the constraint to be syntactically in the structure
binding, as in

	structure S: SIG = ...

or would you only require that the components of the structure have
been constrained at some point?  Consider the following.

* structure T: SIG = ...
  structure S = T

* structure T: SIG = ...
  structure S = T.U

* structure T: SIG = ...
  structure S = struct open T end

* structure T: SIG = ...
  structure S = struct open T.U end

* structure S = struct ... end : SIG

* structure S = let ... in struct ... end : SIG end

* functor F (): SIG = ...
  structure S = F ()

* functor F () = struct ... end : SIG
  structure S = F ()

It's not entirely clear to me what you want to do about structure S in
the above situations.