Eliminate packaging dependency

[glenn-jocher]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Major updates to Ultralytics THOP for enhanced PyTorch model analysis 🚀

📊 Key Changes

• Automation in Publishing: Simplified the criteria to publish new versions automatically. Now, every new version increment triggers a publish. 🔄
• Dynamic Versioning: Shifted to dynamic versioning, aligning with the package version directly in the code, removing the need for a separate version file. 🆚
• Improved Package Description: Updated the package description for better clarity on its purpose. 📝
• License Update: Changed the license specification format in the package metadata. 📜
• Dependency Streamlining: Removed an unnecessary dependency (packaging), making the package lighter. 🗑️
• Adaptation to PyTorch Versions: Removed version checks for PyTorch, assuming compatibility with current versions. 🤹
• Enhanced Operation Support: Included support for nn.SyncBatchNorm across all compatible PyTorch versions. 🔩

🎯 Purpose & Impact

• These changes aim to streamline the development and release process, making it easier for the Ultralytics team to maintain and update the package. 🚤
• Dynamic versioning and automatic publishing ensure the users always have access to the latest features and fixes without manual intervention. 🌈
• By updating the description and removing unnecessary dependencies, the package becomes more user-friendly and lightweight, enhancing the user experience. 🎈
• Ensuring broader operation support and compatibility with current PyTorch versions allows users to analyze more comprehensive models without worrying about manual compatibility checks. 🧰

This update solidifies Ultralytics THOP as a go-to tool for analyzing PyTorch models, making it more accessible, user-friendly, and up-to-date with current PyTorch ecosystems.