nullary _ffi's

[Matthew Fluet]

> Thanks for the patch.  It fixed the problem with the regression failure
> (cmdline).  Unfortunately, it introduced two new -native regression failures,
> bytechar and string.  Do you see those failures as well?

I don't recall having seen any problems with those tests after I patched
the nullary _ffi's, and my current version of MLton doesn't exhibit any
problems.  But I've done a bit of work on the simplify phase and also
added in the overflow checking (although that shouldn't affect the normal
compiles).

Could you post your current snapshot and I'll take a look at it?  I'm
guessing that there have been some changes to the basis-library if it was
the cmdline regression test that prompted the patch to the nullary
_ffi's.

It's probably also a good time to integrate my latest changes.