Update torch

[genevera]

Update PyTorch to CUDA 13.0 Support with Latest Dependencies

🎯 Summary

This PR upgrades the horde-worker-reGen worker codebase from PyTorch cu128 (CUDA 12.4) to cu130 (CUDA 13.1), along with Python version bump and refreshed dependency requirements for improved stability and compatibility with newer GPU hardware.

🔧 What's Changed

• PyTorch Upgrade: Migrated from CUDA 12.4 (cu128) to CUDA 13.1 (cu130) across all build configurations
• Python Version: Updated default Python version from 3.11 to 3.12 in Docker builds
• New requirements.txt: Added properly structured pip requirements with explicit PyTorch installation instructions and additional dependencies (certifi, babel) for SSL and locale support
• entrypoint.sh permissions: Fixed executable bit set on the entrypoint script
• Dependency updates: Refreshed horde packages to their latest versions:
  • horde_sdk~=0.17.1
  • horde_safety~=0.2.3
  • horde_engine~=2.20.12
  • horde_model_reference>=0.9.2
• Documentation: Updated README files to reflect new installation commands and build arguments

📦 Files Modified (10 total)

File	Changes
Dockerfiles/Dockerfile.cuda	CUDA 13.1, Python 3.12, added requirements.txt COPY instruction
Dockerfiles/README.md	Updated build argument docs to reflect cu130
Dockerfiles/compose.cuda.yaml	Updated compose args for new versions
Dockerfiles/entrypoint.sh	Fixed executable permissions, updated pip install command
Dockerfiles/requirements.txt	New file with explicit torch/torchvision deps and requirements
README_advanced.md	Updated installation instructions for cu130
requirements.txt	Restructured with --extra-index-url directive
tox.ini	Updated to use cu130 index URL
update-runtime.cmd	Windows runtime script updated to cu130
update-runtime.sh	Linux/Mac runtime script updated to cu130

✨ Benefits

1. Better GPU support: CUDA 13.1 provides improved compute capability for newer NVIDIA GPUs (including RTX 40-series and future hardware)
2. Python 3.12 advantages: Performance improvements, better error messages, and access to latest language features
3. Explicit dependencies: The new requirements.txt structure makes dependency resolution more reliable
4. Cleaner installs: Removed specific torch version pin (torch==2.9.1) in favor of direct pip install with index URL for flexibility
5. Updated SDK versions: Latest horde packages ensure compatibility and bug fixes

🧪 Testing Considerations

• Build Docker images successfully on CUDA 13.1-compatible systems
• Verify GPU detection within containers (nvidia-smi, torch.cuda.is_available())
• Confirm worker registration and job processing works with new dependencies
• Test existing custom workers/nodes to ensure compatibility with updated horde packages

⚠️ Notes for Reviewers

• The shift from pinned (--==2.9.1) to unpinned PyTorch installation may affect reproducibility considerations - discuss if version pinning is preferred
• Python 3.12 upgrade should be non-breaking but requires verification against any legacy dependencies
• All cu130 URLs use the official PyTorch index: https://download.pytorch.org/whl/cu130
  updates across the horde ecosystem.