PATCH: Update libiberty/testsuite/demangle-expected

[DJ Delorie]

> There was never a functional specification for those failed tests.

The creation of a testsuite created one.

> Regardless if libiberty and libstdc++ share the demangler source or
> not, should they have the same output on the same input for those
> failed ones in libiberty? If not, why?

Not always.  The libiberty demangler supports more mangling schemes
than the libstdc++ one, so the outputs will not always be the same.
The case you patched, for example, is not part of the published ABI so
could conceivably be part of some older ABI (in addition to its
current use).  Because of that possibility, we can't just assume that
the testsuite should be the same between the two.