Improve recast performance for .ply files

[dilirity]

Summary

Optimize PLY mesh loading, parallelize shared pipeline stages (normals, BVH, rendering), and harden the PLY parser against malformed files. Builds on the OBJ performance work merged in #13.

What changed

PLY loader rewrite (MeshLoaderPly.cpp)

• Replaced std::ifstream with single fread into a memory buffer -- eliminates millions of tiny I/O calls
• Bulk memcpy for vertex data (binary layout matches rdVec3D)
• Validated header parsing with strtoll and INT_MAX bounds
• Pre-allocation bounds check: rejects files where claimed vertex/face counts exceed actual file size before allocating
• Face index validation: every triangle index checked against [0, m_vertCount) before use
• Integer overflow protection on all size calculations (long long arithmetic, m_triCount <= INT_MAX/3 guard)

Parallel normal calculation (MeshLoaderObj.cpp, MeshLoaderPly.cpp)

• Chunk-based work distribution (4096 triangles per chunk) with std::atomic counter
• Spawns hardware_concurrency - 1 worker threads writing to non-overlapping output ranges
• Applied to both OBJ and PLY loaders

Parallel BVH construction (ChunkyTriMesh.cpp)

• Replaced qsort + C comparators with std::sort + inlined lambdas
• std::execution::par for sorting arrays >= 32K items
• Parallel bounds computation across all cores (same atomic chunk pattern)
• Parallel tree subdivision: pre-computes subtree node counts, spawns left/right on separate threads for the top 4 levels
• maxNodes guard prevents overlap if pre-computed counts exceed allocation

VBO rendering (Editor.cpp, EditorInterfaces.cpp)

• Cached display list vertex data uploaded to GPU via VBOs (glBufferData with GL_DYNAMIC_DRAW)
• Each frame binds and draws from GPU memory instead of transferring from CPU RAM
• GL buffer extension function pointers loaded via SDL_GL_GetProcAddress with null validation at startup
• VBOs properly created, re-uploaded on cache dirty, and deleted in destructor

Navmesh cache invalidation (Editor.cpp, Editor_SoloMesh.cpp, Editor_TempObstacles.cpp, NavMeshPruneTool.cpp)

• Added invalidateNavMeshCache() calls after mesh rebuild, navmesh load, traverse link rebuild, and static pathing rebuild

How to test

1. Open the NavEditor
2. Load a large PLY mesh (1GB+)
3. Verify the mesh loads and renders at interactive frame rates (expect 60-90 FPS depending on GPU)
4. Build a navmesh and verify it renders correctly
5. Toggle "NavMesh" checkbox off/on -- cache should update without stalls
6. Click "Rebuild Static Pathing Data" -- navmesh display should refresh
7. Try loading a small/malformed PLY file -- should fail gracefully without crashing