Sequence/Parallel filterset

WIP fixes to deal with larger than LLM's context size filtersets request.

Includes:

• create LLM call chain per query/filter pair
• experimental parallel execution of the chains (works with OpenAI gpt-4o, needs fixes for Groq Llama 3.1)
• experimental sequence execution of the chains (works with the in-house LLM endpoint running Hermes 2 Pro Mistral 8B)
• system prompt with instructions
• potential insert point for dynamic few-shot examples