[Bug]: register_for_llm overwrites client, breaking custom model clients

[Basekill]

Describe the bug

When trying to register a function for an LLM, it will overwrite the custom model client registered, leading to a PlaceholderClient.

autogen/autogen/agentchat/conversable_agent.py

Line 2496 in d461100

self.client = OpenAIWrapper(**self.llm_config)

This should not overwrite existing custom model clients

Steps to reproduce

assistant = ConversableAgent(
    name="Assistant",
    system_message="You are a helpful AI assistant. "
                   "You can help with simple calculations."
                   "Return 'TERMINATE' when the task is done.",
    llm_config={"config_list": [custom_config]},
)

assistant.register_model_client(model_client_cls=CustomModelClient)  # Registers custom model client
assistant.register_for_llm(name="calculator", description="A simple calculator")(calculator)  # Overwrites custom model client

1. Create an agent
2. Register a custom model client
3. Register a function for an LLM
4. The client is overwritten so we are left with a PlaceholderClient instead of the registered custom model client

Model Used

Custom Model (with GPT-4)

Expected Behavior

The function should be registered without overwriting the custom model client

Screenshots and logs

No response

Additional Information

No response

[ekzhu]

@Basekill would you be interested in helping out here?