Splitting model in seperate classes

[pescheck-bram]

This PR refactors the model selection and handling logic by splitting the model manager into provider-specific classes, aiming to reduce complexity and improve modularity.

What’s changed

• Created dedicated provider classes:

  • openai_provider.py
  • azure_openai_provider.py
  • anthropic_provider.py
  • deepseek_provider.py

• Introduced a base.py with a common provider interface (BaseProvider)

• Added a registry.py to dynamically resolve and initialize providers

• Updated model_manager.py to delegate model usage to registered provider classes instead of conditionals

• Updated all references accordingly to use the new structure

Benefits

• Eliminates if/else or match logic for provider selection
• Makes it easier to add new providers (just implement a new class and register it)
• Improves readability, testability, and long-term maintainability