Fortran sizetype fix

[Richard Kenner]

    This patch seems to have disappeared into the black lagoon.  It fixes
    some testsuite failures in Fortran.  It's quite short but I would
    appreciate if someone who actually _understands_ the sizetype changes
    would review it.

I certainly understand the sizetype changes, having written them, but
what I can't figure out is that code in the Fortran front end.  I
can't figure out what the calls *expect* *offset to be!  In at least
one case, it's being assigned to a variable which is also being set in
another case to size_zero_node.  If that's correct (and he distinction between
sizetype and bitsizetype predates my changes by quite a while), then your
patch is wrong.  If that's incorrect, then your patch needs to change it too.
But I can't tell if that's wrong or not.