Support for Azure OpenAI

[fschuh]

DsRAG supports OpenAI, but not its Azure variant.

I was planning to add a PR for this, but there's one sticking point on the Semantic Sectioning code:

def get_structured_document(document_with_line_numbers: str, start_line: int, llm_provider: str, model: str, language: str) 

This would need to be refactored to make room for Azure OpenAI's additional configurations parameters - only the model name isn't enough for Azure.
Perhaps a generic config dictionary could be passed in to this function so it could be a bit more flexible?

Even better if it could use the LLM abstract base class, but I see this wasn't done because of the dependency on Instructor.

[zmccormick7]

We really need to refactor how LLM providers and models are specified. It's a bit of a mess right now, with different methods used in different places. I think extending the LLM abstract base class and using that everywhere would be the best approach, because it's so flexible. I believe the structured output part (which is what we're using Instructor for right now) could be figured out. Instructor could be an implementation detail that lives inside specific LLM subclasses, but isn't required.