Fix Behaviour when LLM is None

[CalebCourier]

Previously, if the llm_api argument was None and OPENAI_API_KEY environment variable was not set, the llm_api would match the static openai completions create function because it was returning None. This would cause issues as shown in #979

We now avoid this issue by:

1. Not matching to an LLM if the llm_api argument is None and there is no model argument specified.
2. Not returning None from the static openai fetcher methods. If auth is not properly set, let the OpenAI errors propagate up accordingly.