bug in IntInf.scan StringCvt.HEX

[Stephen Weeks]

The reason I cc'ed MLton on this is that this behavior causes a bug in 
the SML/NJ compiled version of MLton.  MLton uses IntInf in one place
in its implementation: atoms/small-int-inf.fun.  This module converts
strings representing IntInfs in the input program into word32s if the
IntInf is small enough to fit in MLton's represenation of small
IntInfs.  As a consequence of the NJ bug, the SML/NJ compiled version
of MLton does not correctly handle IntInf constants in the input
program when those constants begin with "0x".  I am not sure if this
affects the self-compiled version or not.  I am investigating.