Add non-Ollama embedding support

[TanGentleman]

I want to be able to use LM Studio for both the LLM and the Embeddings. If I understand it correctly, then this would affect how I handle the vectorstore/retriever. I'd have to create a lower level embeddings implementation, which means I can still create a vectorstore from documents, but I'd have to pass the embeddings in directly when performing a similarity search. Seems like this is how it might work:

1. Create vectorstore without an embedding function (is this supported?)
2. Pass user query string to LMStudio Embeddings API. This should return a List[float]
3. Perform a similarity search different from the current retriever implementation, since the embeddings have to be passed directly.

Here's a snippet from the vectorstore code in langchain_core. This is likely what would work best for such tasks.

vectorstore.similarity_search_by_vector(embedding: List[float], k: int = 4, **kwargs: Any
    ) -> List[Document]:

[TanGentleman]

Still having issues with using the LMStudio Embedding API in the same retriever. Going to deprioritize this since it'll likely evolve after it has an official release.