v0.4.1: Add r2SCAN model support and update documentation

What's New in v0.4.1

🚀 New Features

• r2SCAN Model Support: Added support for loading r2SCAN transfer learning models via CHGNet.load('r2scan')
• Enhanced Documentation: Updated README with detailed pretrained models section

🔧 Improvements

• Model Loading: CHGNet.load() now supports:
  • CHGNet.load() - Latest MPtrj-pretrained CHGNet (0.3.0)
  • CHGNet.load('0.2.0') - Deprecated MPtrj version for backward compatibility
  • CHGNet.load('r2scan') - R2SCAN level model transfer learned from MP-R2SCAN dataset
• CI Improvements: Fixed wandb hostname issues in GitHub Actions
• Documentation: Added code examples and references to MatGL/MatPES models

📚 Documentation Updates

• Added comprehensive pretrained models section to README
• Included usage examples for different model versions
• Added reference to MatGL/MatPES models for non-ground-state calculations
• r2SCAN Model Details: See r2SCAN model documentation for detailed information about the R2SCAN transfer learning model

🐛 Bug Fixes

• Fixed wandb hostname length issues in CI environment
• Improved error handling in model loading functions
• Fixed r2SCAN model is_intensive attribute: Set to True for correct behavior

📖 r2SCAN Model Information

The r2SCAN model is a transfer learning model fine-tuned from CHGNet v0.3.0 on the MP-r2SCAN dataset. For detailed information about model parameters, training configuration, and performance metrics, please refer to the r2SCAN model documentation.