Release v0.7.8

Features

• Unified Generic BVH Traversal Engine (bvh_traverse.cuh):
  • Replaced ad-hoc stack management loops across linear BVH queries (bvh.cu, mesh_bvh.cu, triangle_mesh.cu, gs_aabb.cu, and pgs_aabb.cu) with an inlined, generic stack descent engine parameterized with node test and leaf visit functor lambdas.
• Shared Volumetric Extraction & Flood-Fill Utilities:
  • Unified voxel corner sign classification into voxel_classify.cuh, eliminating code duplication across mc.cu and mtg.cu.
  • Consolidated byte-level atomic operations (atomicCAS_int8) and ray-triangle intersection traversal into flood_fill_common.cuh for both single-level and coarse-fine GPU flood fill algorithms.

Refactors & Build

• Kernel Deduplication:
  • Removed over 1,000 lines of duplicated traversal and classification code while maintaining identical compiled machine code and performance.
• Package Version: Bumped version to v0.7.8.

Examples

• High-throughput BVH spatial queries and volumetric isosurface extraction across Gaussian splats, meshes, and SDF volumes.