Add support for separate output resolution model

[kugesan1105]

Description

Implements dual-model processing capability in MTLLM by introducing 'resolve_with' parameter.
This allows output resolution to be handled by a different LLM than the one performing
the primary inference.

Example usage:

predict_vehicle_damage(img: Image) -> DamageType by llm(
    is_custom=True,
    resolve_with=llm2
);