Consolidate init_seeds()

[glenn-jocher]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refined random seed initialization for better reproducibility and performance control.

📊 Key Changes

• Removed init_torch_seeds function from utils/torch_utils.py.
• Included random seed initialization directly in init_seeds within utils/general.py.
• Added control over CUDA's cuDNN benchmarking and determinism settings based on the seed value.

🎯 Purpose & Impact

• Ensures that users have consistent behavior when setting the seed for reproducibility purposes.
• By adjusting cuDNN settings, users can now choose between faster execution with less reproducibility and slower execution with more reproducibility, useful for debugging or production environments.
• Simplifies the codebase by removing redundant functions, potentially making it easier for users to understand and maintain the initialization process.