feat(v5-M4): industrial formats + GPU acceleration + Python bindings

M4.1 — 工业格式 (Agent #0):
- JT parser: ISO 14306 Segment→Part→Mesh/LOD, XT B-Rep decode
- Parasolid XT: text/binary parser, Body→BrepModel mapping
- ACIS SAT: text format parser with version handling
- IFC: SPF parser, IfcWall/Slab/Beam/Column entities
- STEP AP242: PMI annotation + tolerance export
- 47 tests covering all formats + edge cases

M4.2 — GPU 加速 (Agent #1):
- CUDA kernels: mc_kernel, qem_cost_kernel, tri_intersect_kernel
- __constant__ memory for edge tables, atomic triangle collection
- CPU fallback when CUDA unavailable (seamless degradation)
- VDE_USE_CUDA CMake option, .cu compilation support
- 9 tests with CPU path validation

M4.3 — Python 绑定 (Agent #2):
- vde_brep: BrepModel, make_*, boolean, STEP/IGES, heal, validate
- vde_sdf: primitives, operations, to_mesh, gradient descent
- vde_cam: roughing/finishing/drilling, Tool, ToolLibrary
- vde_assembly: Assembly, AssemblyNode, interference, explode
- Version: 3.3.0 synced across pyproject/setup/__init__

13 files, ~5200 lines, 56+ tests

tests/CMakeLists.txt

@@ -16,3 +16,4 @@ add_subdirectory(sdf)
 add_subdirectory(sketch)
 add_subdirectory(foundation)
 add_subdirectory(fuzz)
+add_subdirectory(gpu)