vectorcall kwnames C array #699

So vectorcall protocol allows more efficient calling without needing to construct Python containers.

Still, it needs a tuple for knames.

I am modifying keyword arguments (functools.partial), where I sometimes need to modify kwnames and this results in big performance penalty when this is the case.

Not sure if this could result in more general performance improvement by not having to construct tuple in other places, but maybe this would be worthwhile considering?

E.g.

vectorcall(
    partialobject *pto,
    PyObject *const *args, size_t nargsf, 
    PyObject *const *kwnames, Py_ssize_t nkwds)

Or something similar.

Will look into this in due time, but thought maybe it is worth posting to get some thoughts.