enabler: decide on strategy for local LLMs/embeddings

[ludwiktrammer]

The aim of this ticket is to prepare a document with Ragbit's strategy when it comes to running LLMs/embeddings locally:

Do we want to directly support running models locally? If yes - using which libraries. If no - what's the alternative (for example: documenting how to use LiteLLM to connect with local models available via http API)