switched to qwen

Updated the chat UI and backend so MiniCPM thinking renders

MiniCPM5 ZeroGPU migration

fix empty research findings