Test for mprotect failure in dl-load.c (bug 12492)

[Roland McGrath]

That's fine.  My first inclination was to say the message should say what
it was doing mprotect for.  OTOH, you're reusing an existing message (also
vague, though previously unambiguous if you looked in the source) so no new
translation is needed.  There remain three mprotect calls in dl-load.c that
don't check for errors.  But since those are reducing access rather than
increasing it, perhaps that's OK.