upgrade_pythoncapi.py: Wrong transformations for Py_Is

Consider an example:

if (((PyObject*)obj)->data == Py_None) {
    111;
}
if (Py_None == ((PyObject*)obj)->data) {
    111;
}

After upgrade_pythoncapi.py I got:

#include "pythoncapi_compat.h"

if (((PyObject*)obj)->Py_IsNone(data)) {
    111;
}
if (Py_None == ((PyObject*)obj)->data) {
    111;
}

Real world example:
https://github.com/aleaxit/gmpy/blob/eb8dfcbd84abcfcb36b4adcb0d5c6d050731dd75/src/gmpy2_xmpz_misc.c#L237

I'm not sure if this is a bug. This kind of issues is not easy to avoid, using regexps for code transformations. Have you considered to use something like the pycparser?

Thanks for the project.

Start with upgrade_pythoncapi.py and reproduce both C examples from the issue, then compare the output with the referenced gmpy2_xmpz_misc.c case. Check whether the transformation handles both operand orders without corrupting member access; done means valid C is produced with the intended Py_IsNone transformation.