Increase max_new_tokens default to 1024 to prevent truncation

Update handler: Remove Python code generation, enforce IB Math tutor style

Add custom handler for Inference Endpoints