GFLOPs computation fix for classification models

[glenn-jocher]

Improved robustness in reading input channel count.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improvements in model pruning and information utilities.

📊 Key Changes

• Replaced a print statement with a LOGGER.info call for model pruning feedback.
• Made the model pruning step permanent within the same function.
• Refactored model_info function to use imgsz as the argument name instead of img_size.
• Updated FLOPs calculation in model_info with more concise and efficient code.
• Changed the FLOPs calculation to use a copied model instance, providing stride and image size flexibility.

🎯 Purpose & Impact

• Enhances codebase consistency and compliance with standardized logging.
• Simplifies the model pruning process from a user standpoint.
• Improves code readability and maintainability by unifying argument nomenclature.
• Optimizes performance analysis calculations to better estimate computational requirements.
• Users will benefit from clearer reporting on model pruning and a more accurate calculation of computational costs related to model complexity and inference speed.