[REFACTOR] Decouple provider configuration to support multi-provider architecture

[thomas-vilte]

Description

I have observed that the current configuration struct and initialization logic are tightly coupled to the Google Gemini provider. The existing codebase assumes a single active provider and maps configuration fields directly to Gemini-specific parameters. We need to refactor the configuration package to support a generic Provider interface or a map of provider configurations. This change will allow the system to load settings for Ollama, OpenAI, and Gemini simultaneously.

Reason

We should implement this refactor to improve the maintainability and extensibility of the codebase. As we plan to support local LLMs via Ollama and other cloud providers like OpenAI, the current rigid configuration structure acts as a blocker. Decoupling the config now will simplify the implementation of dynamic provider switching and fallback mechanisms in future releases.