GeezerVison

This wasn't simple after all. A test crashed because there was an error path with a manual goto error when the inputs had already been decref'ed. This caused a reference deficiency for the inputs on the stack leading to a crash that had me stumped until I picked up paper and pencil and reasoned through the generated code.

While I think I've fixed this now, this points out a general weakness in our bytecode definitions. Sometimes we feel forced to leave a manual goto error in the code (though note that in this case it was an oversight, for which I take full responsibility), but maintaining reference count correctness in all cases is hard!

🤖 New build scheduled with the buildbot fleet by @gvanrossum for commit 1e48f9d 🤖

If you want to schedule another build, you need to add the :hammer: test-with-refleak-buildbots label again.

Ready for review.

This is blocked until I can figure out the refleak failures. (EDIT: But first I'm going to re-run that particular build, in case it's a flake.)

Hm, aarch64 RHEL8 Refleaks PR seems to have been failing a lot, historically. Ignoring.