Thanks to Stefan Behnel <stefan_ml@behnel.de> for pointing this out.

but we can get away with skipping intermediate classes if nothing new happens

The signatures are no longer required to match if the only argument to __cinit__ is self.

GetStarArgs: separate out keyword handling, generate specific code for functions with or without any combination of *args and **kwargs

split up __Pyx_GetStarArgs() into two separate functions for *args and **kwargs to prepare their independent use

use PyObject_Call() instead of _CallObject() just as Python calls it internally, speedup of about 5%