bpo-46881: Fix refleak from GH-31616

[JelleZijlstra]

@Fidget-Spinner found the leak and the patch but asked me to submit
it.

Confirmed that this fixes the leak with ./python.exe -m test test_sax -R 3:2

https://bugs.python.org/issue46881

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @JelleZijlstra for commit 92ecb62 🤖

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

[Fidget-Spinner]

Thanks Jelle!

[vstinner]

About the commit message:

[bpo-46881](https://bugs.python.org/issue46881): Fix refleak from GH-31616 (GH-31805)

Next time, would you mind to elaborate the commit message? What was leaking? What is the change about? Why does it leak? etc.

I recently looked for that commit and just from its description (without going to bpo-46881), it's hard to guess what the change does.

I'm trying to make the assumption that the developer is offline (in a train, an airplane or a boat) and cannot access the bug tracker. By the way, bugs.python.org may disappear one day. Python infra is changing "often" if you look at Python history: CVS is gone, SVN is gone (SVN commit numbers cannot be easily used), Mercurial server is still up, now we use Git. Bugs were on SourceForge, are on Roundup but are moving to GitHub :-)

Note: I do actually work often in trains these days with unusable wifi/4G :-)

[JelleZijlstra]

Thanks for the feedback! I looked at other recent refleak fixes in git log and usually they at least mention the file or function where the refleak is. I'll try to do that next time.