[ENH] Add PyTorch backend support for dual contouring and octree operations

[Leguark]

[ENH] Adding PyTorch GPU Support

This PR adds support for PyTorch GPU acceleration by:

• Implementing a PyTorch backend for tensor operations throughout the codebase
• Adding GPU device detection and configuration in the backend tensor module
• Creating PyTorch-compatible implementations of key functions like packbits, to_numpy, and matrix operations
• Adapting octree generation and dual contouring algorithms to work with PyTorch tensors
• Ensuring proper tensor conversion between CPU and GPU when needed
• Implementing tensor-specific operations for both backends to maintain compatibility
• Adding proper memory management for contiguous arrays in PyTorch

The implementation ensures that operations can run on either CPU or GPU when using the PyTorch backend, with appropriate error handling when CUDA is not available but GPU is requested.

[Leguark]

• [FIX] Reuse weights calculation instead of raising error on hash mismatch #27
• [ENH] Refactor dual contouring with parallel and sequential triangulation modules #26
• [ENH] Refactor dual contouring code with improved documentation and modular functions #25
• [REFACTOR] Use center grid for dual contouring instead of corners grid #24
• [FIX] Use timestamp for cache key instead of input parameters #23
• [FIX] Fix PyKeops tensor operations and ensure contiguous arrays #22
• [ENH] Add PyTorch backend support for dual contouring and octree operations #21 👈 (View in Graphite)
• [!!BUG] Chunking with faults were creating rogue voxels | GEN-13299 #20
• [ENH] Add transform support to meshes_to_subsurface method #19
• [FIX] Separate pykeops_enabled flag from use_pykeops for better control #18
• [ENH] Improve CG solver and add weights caching for better performance #17
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.