Refactoring suggestions

[MichaelOwenDyer]

I have some suggestions for the LLM subsystem code. @Hialus, not trying to step on your toes, every change in this PR is a personal opinion. Could be that I am missing the bigger picture with some of these things, let me know your thoughts. Here's a summary of what I did:

• Renamed classes to reflect their conceptual purpose rather than implementation-oriented. In my opinion, conceptual class names make it easier to reason about the code using those classes.
• Took out some code in basic_request_handler that I thought was redundant.
• In LLMManager, I replaced the linear LLM search with a dictionary lookup by ID. Even though O(n) would not be a performance bottleneck with the low amount of LLMs we have, let's do it in O(1).
• Renamed the methods completion(), chat_completion(), create_embedding() to complete(), chat(), and embed() for consistency
• Added one requirement for PyYAML since it is being used in LLMManager, and changed the requirements in requirements.txt to use ~= instead of ==.

[Hialus]

As it stands right now you did a lot of changes in the internals of the LLM subsystem. I very clearly stated that the entire internals of the subsystem are subject to change and further improvements. The current implementation is merely a simple first version to use in the development of the pipeline subsystem.

The changes to the RequestHandler are sensible and I'd suggest we implement them, as we already discussed that.
The changes to the LlmManager are entirely unnecessary, as the current implementation will effectively be completely replaced. As it is not part of the public API of the subsystem this is easily doable. So any changes here do not really make much sense.
For the changes about the AbstractLlmWrapper I'd need to spent some additional time thinking about the best possible name, but as they are internal classes this is not really high priority and can be done at any time.

Also your code fails the linter checks. I'd suggest to install the pre commit hooks.