let you get away with typeid(T) when T is an incomplete class
type (that's conforming behavior), but GCC at least will allow
typeid(T&) -- also disallowed by the standard when T is incomplete.
If it turns out that EDGs also barf on typeid(T&), we may have more
work to do.
Some warning suppression for MSVC.
[SVN r29020]
Fix wstring support so it doesn't break gcc2.95.x, which has no wstring.
Modify CallPolicies protocol so it can safely adjust the argument tuple.
[SVN r20090]