REF: Use chat_template for LLM instead of prompt_style

[ChengjieLi28]

1. Remove all the prompt_style related codes
2. Use messages instead of prompt, system_prompt and chat_history. which is compatible with OpenAI API.
3. Remove qwen-chat tool call function.
4. Remove qwen react tool call. Use normal tool call instead. See 工具调用chat_template格式咨询 QwenLM/Qwen2.5#805
5. Now register function is recommanded for registering the LLM models with generate and chat abilities that not in the builtin list.
6. Registering models for tool call and vision models must be the builtin model families.
7. Fix: the last one of streaming results of LLM is not match the OpenAI format.
8. Fixes 自定义模型 xuanyuan-6b 模型，模型推理时无法正确停止。 #1940
9. TODO: web UI
10. TODO: doc

[qinxuye]

LGTM