newlib bug in argz_insert

[Eric Blake]

I discovered and patched a newlib bug in argz_insert today
(http://sources.redhat.com/ml/newlib/2005/msg00516.html):
when the before argument was NULL, it was dying with EINVAL
rather than correctly appending to the end.  Is it worth updating
m4/argz.m4 to detect this bug on systems using old newlib, and
find if any other system has the same bug, or do we just continue
with the status quo and let detection of a system argz_insert
avoid instantiation of the gnulib version, even if the system
version is buggy?

--
Eric Blake