feat: add X-ASR + Qwen3-ASR backends, Gemma-4 LLMs; fix review findings

Add speaker name detection feature

Fix Unicode encoding in export headers for Chinese filenames

re-implement in FastAPI