convert sep_style to int raise ValueError in some case during base_model_worker init

Nobody has claimed this yet.

* Environment
* Python 3.11.5
* FastChat 0.2.35
* llm model: gpt-3.5-turbo
* Error logs
```
2024-01-31 20:35:54 | ERROR | stderr | Process model_worker - gpt-3.5-turbo:
2024-01-31 20:35:54 | ERROR | stderr | Traceback (most recent call last):
2024-01-31 20:35:54 | ERROR | stderr | File "/root/miniconda3/envs/codefuse/lib/python3.11/multiprocessing/process.py", line 314, in _bootstrap
2024-01-31 20:35:54 | ERROR | stderr | self.run()
2024-01-31 20:35:54 | ERROR | stderr | File "/root/miniconda3/envs/codefuse/lib/python3.11/multiprocessing/process.py", line 108, in run
2024-01-31 20:35:54 | ERROR | stderr | self._target(*self._args, **self._kwargs)
2024-01-31 20:35:54 | ERROR | stderr | File "/root/codefuse-chatbot/examples/llm_api.py", line 564, in run_model_worker
2024-01-31 20:35:54 | ERROR | stderr | app = create_model_worker_app(log_level=log_level, **kwargs)
2024-01-31 20:35:54 | ERROR | stderr | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-01-31 20:35:54 | ERROR | stderr | File "/root/codefuse-chatbot/examples/llm_api.py", line 392, in create_model_worker_app
2024-01-31 20:35:54 | ERROR | stderr | worker = ModelWorker(
2024-01-31 20:35:54 | ERROR | stderr | ^^^^^^^^^^^^
2024-01-31 20:35:54 | ERROR | stderr | File "/root/miniconda3/envs/codefuse/lib/python3.11/site-packages/fastchat/serve/model_worker.py", line 66, in __init__
2024-01-31 20:35:54 | ERROR | stderr | super().__init__(
2024-01-31 20:35:54 | ERROR | stderr | File "/root/miniconda3/envs/codefuse/lib/python3.11/site-packages/fastchat/serve/base_model_worker.py", line 48, in __init__
2024-01-31 20:35:54 | ERROR | stderr | self.conv.sep_style = int(self.conv.sep_style)
2024-01-31 20:35:54 | ERROR | stderr | ^^^^^^^^^^^^^^^^^^^^^^^^
```
* When i debug the code, i found that in `conversation.py`, function register_conv_template for ChatGPT (as same as Perplexity ai and Bard) sets the sep_style to None by default, which is the root cause the Error above
```
# ChatGPT default template
register_conv_template(
Conversation(
name="chatgpt",
system_message="You are a helpful assistant.",
roles=("user", "assistant"),
sep_style=None,
sep=None,
)
)

# Perplexity AI template
register_conv_template(
Conversation(
name="pplxai",
system_message="Be precise and concise.",
roles=("user", "assistant"),
sep_style=None,
sep=None,
)
)
```

Is it a bug needs to be fixed?

Start with conversation.py and the register_conv_template definitions for ChatGPT and Perplexity AI, then trace the initialization in base_model_worker.py where sep_style is converted to an integer. Check how the model worker in examples/llm_api.py supplies these templates. Done means initialization no longer raises ValueError for templates whose sep_style is None, with the intended template behavior preserved.

Written by the indexing model from the issue text.