Fix Custom LLM Support

[CalebCourier]

Fixes: #979

This PR has two main focuses:

1. Fix support for custom llm_api callables
   • Fix and safeguard when checking if the llm_api is one of OpenAI's static functions
   • Root cause of issue, if OPENAI_API_KEY is not in the environment nor set in an openai global, openai will raise an exception when you try to reference any of the static completion functions even if you're not calling them. We were not properly handling this exception. Now we do.
2. Method signature enforcement for custom llm_api callables
   • In order to give users upfront feedback, we now inspect custom llm_api callables to ensure they allow for a positional argument for the prompt and accept a spread/rest variable for keyword arguments.
   • We also warn if we do not see keyword arguments for instructions or msg_history since not destructuring out can have unintended consequences when passing **kwargs to any downstream code.

We've also increased test coverage to catch any future regressions to custom llm_api support.

TODO

• Doc on custom llm_api callables