[MLton] IntInf implementation questions

[Matthew Fluet]

I have a few questions/suggestions for the IntInf implemenation:

1) Why does the fill function in IntInf.c require space for 2 limbs,
    rather than just space for 1 limb?  It seems that we only use one
    limb.

2) Is there a reason that IntInf_{quot,rem} have such complicated
    implementations, rather than simply using mpz_tdiv_{q,r}?  Those GMP
    functions appear to have the right semantics.

3) Why does the bigMul function in int-inf.sml make use of an _import-ed
    multiplication function for small multiplication?  Wouldn't it be more
    efficient to perform the multiplication with Int.* and handle the
    overflow?  Even on the C-codegen, overflow checking primitives are
    inlined, and with the x86-codegen I see a considerable speed up on an
    IntInf.int factorial benchmark that is dominated by small
    multiplications.