Tested 595.58.03 driver. The issue is still present, but seems to be significantly less prevalent.
It’s almost bearable at this point, but still quite distracting.
I would really appreciate any pointers about debugging this. Thankfully, the kernel modules are open-source, so I can tweak them and recompile in attempts to diagnose the issue. But the codebase is enormous and I don’t even know where to start (assuming the problem is in the open parts anyways). Even the diff between last-good and first-bad versions is something like 200k lines.