[MLton] implement _address and _symbol

[Matthew Fluet]

> I guess you are correct: if you want to export an ML object in a C variable
> then it is going to need the Pinned stuff from the ML side.
> Are we in agreement that for parameters however, the pinning is going to
> happen in the C code, not in the _import types?

I think you definitely need a pinning API on both the ML side and the C 
side.  I could still see situations where it makes sense to pin an ML 
object on the ML side before passing it to multiple C functions, avoiding 
the overhead of pinning/unpinning on the C side.