Improve callback performance (C to Python calls)

[rogerbinns]

There are two things that can be done:

• Cache a PyUnicode object for the string of the callback name instead of looking it up each time
• Use the vector calling protocol

A quick benchmark where only the xColumn callback was changed to use PyObject_VectorcallMethod, measured in calls per second by vtbench

2,085,699 Base (Old calling, no string cache)
2,253,161 Vector, PyUnicode_FromString called each time
2,678,927 Vector, PyUnicode_FromString called once
Update after initial implementation has 3 million calls per second.

• vtable vector calls
• vfs vector call
• Other Call_PythonMethod sites
• PyObject_Call sites
• getfunctionargs needs vargs rewrite
• Make it work on Python 3.8
• Memory leak checking
• Connection.execute{,many} use fastcall
• PyObject_Get/SetAttrString - use string table