Hi all, just a gentle follow-up on this PR. I'd appreciate any feedback when time permits. Let me know if there's anything I can improve or clarify. Thanks!

Thanks for the feedback! I think we can test this using image comparison—by generating a target image and validating output against it.

On a related note, I'm not entirely confident that my current method for checking whether the renderer is a vectorized backend is the most robust or appropriate approach. I'd really appreciate any input or suggestions on improving that part of the implementation.

I suspect whether the backend has vector output is not actually the right test here; indeed, mplcairo doesn't exhibit the#28549 bug even for vector output. Rather, the bug likely resides either in backend_{pdf,ps,svg}, or, more likely (just my guess), in MixedModeRenderer.
Therefore, I think I'd prefer if we could figure out what's wrong in MixedModeRenderer (if anything) and directly fix it there, or, if not, at least just directly test whether the renderer is a MixedModeRenderer (_is_mixedmode_renderer) and run the extra code in that case.