Q: class name hashes

[Leopold Toetsch]

We currently seem to have two different hashes for storing class names: 
one for PMC base classes and one for dynamic PMCs and objects.

I think both should use just interpreter->class_hash. OTOH putting PMC 
names into this one hash makes it more likely that we get name 
collisions for user class names.

We should be able to subclass PMCs as real objects.
b3.py seems to do something like:
# class TT(T):
   getclass $P0, "Integer"   # Integer PMC class, Python: "int" class
   subclass $P1, $P0, "TT"   # subclass the PMC
Then it redefines the "__repr__" method, which we BTW don't have. Using 
"get_string" instead could be a problem. Seems that we need a C<repr> 
vtable, which gets called when printing PMCs.

Comments welcome,
leo