switch to using PySet_GET_SIZE() for set size and truth requests as it's fast and also works as expected in PyPy

Add declarations for PyObject_Malloc()

change declaration of GC related type slot functions to use 'PyObject*' instead of 'object' to prevent accidental reference counting

optimise kwargs dict creation away when it's only being passed on and not otherwise modified (ticket 744)

prevent fallback to absolute cimport when relative cimport is not found; generally clean up relative cimport code