Refactor models

• There are now separate methods for creating messages for chat mode
• Subclasses only take care of calling the model-specific completion methods
• Model calls can be cached: _call_model will take care of it
  • Models are not taken into account for caching directly, but the options and kwargs used to initialize them are
  • Currently using the joblib.Memory module, not sure what the best option for caching (other options include ring, diskcache) is

Hasn't been tested using OpenAI API so far.