[MLton] IntInf implementation questions

[Henry Cejtin]

Re dontInLine, it may not use any primitives, but it certainly relies on a lot
of things about MLton (that recursive functions aren't unrolled, that the
simplitier will get rid of the junk, etc.).

Any way, I quite agree that the hack for small*small->small should be quite a
win.  It really is a shame (for IntInf any way) that the tag for non-pointers
isn't 0, but 1.  Especially for + and -.