OpenAIChatCompletionClient should be a ContextManager

[Leon0402]

Right now OpenAIChatCompletionClient is no Context Manager and instead relies on a deletion method for cleanup. In particular it does:

• Create an AsyncOpenAI client here:

  autogen/python/packages/autogen-ext/src/autogen_ext/models/openai/_openai_client.py

  Line 1017 in 34bc82e

  client = _openai_client_from_config(copied_args)

• This is a subclass of AsyncAPIClient, which creates an AsyncHttpxClientWrapper here: https://github.com/openai/openai-python/blob/main/src/openai/_base_client.py#L1449
• The AsyncHttpxClientWrapper does cleanup in a deletion method here: https://github.com/openai/openai-python/blob/main/src/openai/_base_client.py#L1362

So we rely on __del__ for cleanup of resources, which is not too great. There are many posts about the danger of it, although I am not quite sure what exactly applies to a modern python version as there has been some PEPs improving the situation. Nevertheless, there seems to be a common agreement that context managers are always preferred for cleaning up resources. As AsyncAPIClient already is a context manager, we would just need to make OpenAIChatCompletionClient one as well (an async one to be more concrete).

[ekzhu]

Good point. We may end up having states inside the client that requires clean up. Consider perhaps a close() method for the client class as the first step?