Added Support for Multiple Models

Added

• Provider-agnostic LLM factory boundary with app code depending on shared LLM
  interfaces instead of concrete provider implementations.
• App-owned provider/model registry for coherent OpenAI and Anthropic runtime
  model sets across chat, helper, builder, tagger, and Dream roles.
• Anthropic provider support behind the existing LLM factory and runtime
  routing boundary.

Changed

• Local setup and model-selection documentation now favor registry-driven
  provider defaults over hard-coded model inventory in environment files.
• RAG route score thresholds are tuned to keep retrieval useful while filtering
  weaker candidates.