bpo-46841: Fix BINARY_OP's handling of inline caches

[brandtbucher]

https://bugs.python.org/issue46841

[markshannon]

Changes LGTM.

OOI, how did you find this?

Would it be feasible to add a regression test?

[brandtbucher]

OOI, how did you find this?

After the LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE issue on #31640, I went through both these files and reviewed our use of instruction pointers. #31664 came out of that too.

Would it be feasible to add a regression test?

Not until we can dis quickened code. Once that happens, I can go through and add a new suite of test cases for all of our adaptive/specialized/super instructions.

Don't worry, it will be less like test_dis.py and more like test_peepholer.py. ;)