The use of `PyStackRef_AsPyObjectBorrow` makes it hard to track ownership of references, making analysis of escaping calls too difficult

The problem with PyStackRef_AsPyObjectBorrow is that it is not clear whether the reference is owned by the stack ref or the pointer.

Borrowing references across calls is fine but in less structured code, it is error prone and very hard to analyze.

The solution is change most, ideally all, uses of PyStackRef_AsPyObjectBorrow to PyStackRef_AsPyObjectSteal so that the ownership of the reference is clear.
E.g.

        inst(UNARY_NEGATIVE, (value -- res)) {
            PyObject *val_o = PyStackRef_AsPyObjectBorrow(value);
            PyObject *res_o = PyNumber_Negative(val_o);
            PyStackRef_CLOSE(value);
            ERROR_IF(res_o == NULL, error);
            res = PyStackRef_FromPyObjectSteal(res_o);
        }

would become:

        inst(UNARY_NEGATIVE, (value -- res)) {
            PyObject *val_o = PyStackRef_AsPyObjectSteal(value);
            PyObject *res_o = PyNumber_Negative(val_o);
            Py_DECREF(val_o);
            ERROR_IF(res_o == NULL, error);
            res = PyStackRef_FromPyObjectSteal(res_o);
        }

This ensures that during the escaping call to PyNumber_Negative, the reference to the value a PyObject *, so will not be reclaimed by the garbage collector.

Linked PRs

• gh-122037

CPython のソースで PyStackRef_AsPyObjectBorrow の使用箇所を検索し、issue に示されている PyStackRef_AsPyObjectSteal と PyStackRef_CLOSE の所有権パターンを確認します。どの使用箇所が、参照が外部に渡り得る呼び出しをまたぐのかを特定し、そのうえで関連する参照が有効なままであり、所有権が明示的であることを検証します。リンクされた PR gh-122037 と作業内容を比較します。