[MLton] better support for C pointers

[Stephen Weeks]

> Another idea:
> 
> A malloc / free interface? Using ML-types maybe?

I think it would be nice to have this, but it might be better to base
it on C types that are embedded in phantom types.  Have a look at
Mathhias Blume's NLFFI paper, especially section 3.

	http://people.cs.uchicago.edu/~blume/papers/nlffi.pdf

I think one might be able to attach get/set functions to his RTTI
(run-time type information) to be able to have malloc "know" the right
size.  I haven't done any coding yet though, so I'm not sure.

Also, with the new MLton.Pointer in place, I think we're pretty close
to having enough primitives so that someone could add support for
NLFFI (or an NLFFI-like system) if they wanted.