[FEAT] Support "Hybrid" model selection

[NeonDaniel]

Objective

Currently, Gemini, Ollama, and vLLM are selectable options for the LLM backend to use. A "Hybrid" option should allow for selecting a different backend per-Advisor and for the Orchestrator.

Initial Implementation Requirements

• Update configuration to define a "Default Backend" for each persona
• Add a "Hybrid" option to the selection UI
• Persist Hybrid configuration to user profiles
• Handle user profile config per-advisor at inference time

Other Considerations

• BrainForge personas are determined at runtime; those personas will always need to call the BrainForge endpoint (not configurable to use others)