[MLton] Darwin/PPC changes

[Henry Cejtin]

As I recall (from long ago playing on Mach) the point about _end and _etext
and _edata being problematic was that shared libraries got more random
addresses, but it shouldn't cause a problem for non-shared-library code,
which is all we get under Linux.  The key point is to see if the difference
between get_etext() and the lowest use code address (how you get that varies
with the OS) is close to the code size as printed out by the size command.
If it is, then using that should be fine.