Refactor LLM pipeline to support multiple framework methods

[davidmezzetti]

Currently, txtai primarily supports running local models through the transformers framework. The LLM pipeline should be refactored to support the following additional LLM frameworks.

• Add llama.cpp support to LLM #611
• Integrate with Litellm #554 - adds support for over 100+ LLM APIs
• External generation methods, similar to other components (e.g. ANN, Database, Scoring, Vectors)

These new methods will all be optional. The default method will continue to be local models through Hugging Face Transformers.