Release v0.5.0

Features

• GPU-Accelerated Volumetric Flood Fill (sign_mode=3): Implemented high-performance atomic-queue BFS flood fill kernel to compute topological inside/outside voxel states (Water, Dry, Dam, Collision), eliminating floating interior seams and multi-component self-intersection artifacts.
• Hybrid Fast Winding Number & Pseudonormal Consensus (sign_mode=4): Developed a parameter-free hybrid sign computation architecture that globally unions intersecting geometries via Fast Winding Numbers while dynamically invoking angle-weighted pseudonormals within surface transition zones (0.25 < WN < 0.75).
• Direct Built-In Normal Pre-computation: Upgraded tmesh2sparse and tmesh2voxel to directly trigger compute_triangle_normals(), compute_vertex_normals(1), and compute_edge_normals() on TriangleMesh before chunked queries for zero loop latency.
• PyBind & Data Structure Extensions: Exposed compute_flood_fill operator, flood volume properties, and expanded MeshBVH/TriangleMesh query signatures in C++ and Python bindings to seamlessly orchestrate sign modes 3 and 4.

Refactors & Build

• Pseudonormal Default in Mesh Conversions: Refactored mesh2voxel_sdf to default to angle-weighted pseudonormal dot-products (sign_mode=2) instead of ray casting parity.
• Robust CUDA Build System: Upgraded setup.py to automatically detect and export CUDA_HOME from CONDA_PREFIX and include -allow-unsupported-compiler in NVCC_FLAGS for seamless compatibility with newer Linux system host compilers.
• Version Bump: Upgraded package release version to 0.5.0 in pyproject.toml.

Examples

• SDF Precision Benchmarking: Added an experimental verification benchmark (experiments/test_sdf_libigl.py) evaluating conquer3d sphere SDF calculation precision and GPU throughput against libigl.