That's right this is a nontrivial problem that I struggled with too for gtk-llm-chat! I resolved it using the streaming markdown-it-py library.
Huh this might be another approach with a bit of effort. Thanks for that. I didn't know about this