another limit check bug

[Matthew Fluet]

> > Checked in this fix along with -gc-check option for both c-codegen and
> > x86-codegen. 
> 
> Sorry, I should have mentioned this earlier.  Would it make sense to
> implement gc-check in the backend rather than in the codegens?

I don't think it makes that much of a difference.  At the backend/machine
IL level, we aren't really doing branches or comparisons -- LimitCheck's
are just statements.  I'm happy enough with what's there; considering that
a revision of limit checks and allocation primitives is on the horizon,
I'd vote for letting this stuff stay and save significant changes to limit
checks until the revision.