Release v0.2.8

Features

• Implemented highly optimized GPU Marching Cubes pipeline including C++ bindings, CUDA kernel, and lookup tables.
• Added structured voxel grid creation utilities (C++ and Python wrapper).
• Exported new ops python module containing marching_cubes and create_voxel_grid.

Refactors & Build

• Refactored bindings across all structures (bvh, gs_bvh, kdtree, mesh_bvh, pgs_bvh, triangle_mesh) to improve API consistency.
• Updated core constants and input check macros.
• Cleaned up unused mkdocs patching scripts.
• Resolved significant CI/CD and deployment issues, particularly fixing PyTorch cpp_extension matching with CUDA paths on GitHub Actions.
• Fixed pybind11-stubgen PYTHONPATH execution to ensure robust .pyi stubs generation during PyPI packaging.
• Applied Google-style docstrings across the codebase (TriangleMesh, KDTree, BVH, GSBVH, MeshBVH, PGSBVH).

Examples

• Completely reorganized and rewrote comprehensive markdown descriptions for all 13 Jupyter notebook examples.
• Added comprehensive end-to-end marching_cubes.ipynb demonstrating mesh initialization, implicit field queries, and isosurface extraction.
• Refined 05_query_point_sdf.ipynb with precise benchmarks.