restructure llm organization, to more direct case; split openai/local…

[kuivi]

1. Config Restructuring (config.yml)

   Replaced flat model configuration with structured per-component settings: Before:

   model_type: "openai"
   model_name: "gpt-4.1-nano"
   model_name_rag: "gpt-4.1-nano"
   model_name_tools: "gpt-4.1-nano"
   model_name_combine_agent: "gpt-5-mini"
   model_name_agents: "gpt-4.1-nano"
   After:

   llm_rag:
   model_type: "openai"
   model_name: "gpt-4.1-nano"
   llm_smart:
   model_type: "openai"
   model_name: "gpt-4.1-nano"
   llm_combine:
   model_type: "openai"
   model_name: "gpt-4.1-nano"

2. Code Updates (across 5 files)

   Updated all references to use new nested structure:
   config['model_name'] → config['llm_combine']['model_name']
   config['model_type'] → config['llm_combine']['model_type']
   config['model_name_rag'] → config['llm_rag']['model_name']
   config['model_name_tools'] → config['llm_smart']['model_name']
   Files modified:
   climsight_engine.py (47 lines changed)
   smart_agent.py (20 lines changed)
   streamlit_interface.py (12 lines changed)
   rag.py (2 lines changed)
   evaluation.py (2 lines changed)
   climsight.py (2 lines changed)

[dmpantiu]

This pull request focuses on refactoring how Large Language Models (LLMs) are organized and configured within the climsight codebase. The primary goal is to move from a flat configuration structure to a more modular, "direct case" approach that splits configurations for different tasks (RAG, Smart Agent, and Synthesis).