Normalization should eliminate trivial dependencies on the class name

[fweimer]

The normalization process does not treat class self-references specially (either references to the binary class name, or the name included in nested class names after $ mangling). As a result, the normalization depends on the name of the class. It would be nice if this could be avoided, so that copies of essentially the same class result in the same hash.
It might be interesting to extend this treatment to the package name, but if a copy of the class is created, it might still reference classes in the original package, so neither choice seems to have a clear advantage.