feat: expose progress_callback in Llama wrapper

[Niels-BW]

Exposes progress_callback and progress_callback_user_data through the high-level Llama constructor.

llama_model_params already supports these fields, and the ctype bindings already expose them, but they are not currently available through the Llama API.

This allows Python applications to receive model loading progress and optionally abort model loading by returning False from the callback, consistent with the underlying llama.cpp API.

Implementation

• Add progress_callback and progress_callback_user_data to the Llama constructor
• Wrap the supplied Python callback with llama_progress_callback
• Store the wrapped callback on the Llama instance to prevent it from being garbage collected before model loading completes.

Testing

This change was tested by successfully receiving model loading progress callbacks and aborting model loading by returning False from the callback.