gh-144054: shutdown fix for deferred ref counting

[nascheme]

When shutting down, disable deferred refcounting for all GC objects. It is important to do this also for untracked objects, which before this change were getting missed.

• Issue: Leak with untracked tuples and deferred ref counts #144054

[nascheme]

@colesbury I think this is a correct fix but some of the details are a bit tricky. It seems the !_Py_IsImmortal(op) test is required before calling disable_deferred_refcounting() but I'm not 100% sure why. I'd expect that immortal objects should not also have deferred ref counting enabled for them. The need to check for _PyObject_GC_IS_TRACKED() before calling frame_disable_deferred_refcounting() makes sense.

I split the "disable deferred" into a separate heap scan. Doing it inside scan_heap_visitor() would add extra branches to the hot executate path. So, I think a separate scan and keeping scan_heap_visitor() faster is better.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @nascheme for commit 296eba9 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F144055%2Fmerge

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

[encukou]

The refleak test failure in test_capi (test_replaced_interpreter) is unrelated (#142913)

[encukou]

Otherwise, buildbots are happy.
I'll merge to help fix the tier-1 failures. Any follow-ups can go in separate PRs.