feat(importer): integrate meshoptimizer for vertex cache, overdraw, and LOD

[JeanPhilippeKernel]

Context

The Bistro FBX import produces a 292 MB .zemesh file (2.8M triangles, no optimization). This causes:

• Cold-start hot-reload crash: 292 MB exceeds the 64 MB deserializer scratch
• Slow GPU rendering: vertices are in import order, not GPU-cache order
• Excessive VRAM usage

Proposed solution

Integrate meshoptimizer (https://github.com/zeux/meshoptimizer) — MIT, single header+source, used by Bevy, O3DE, Godot.

Run a 3-pass optimization pipeline in both FbxImporter and GltfImporter after building the flat vertex/index arrays, before SerializeMeshAssetFile:

Pass 1 — Vertex cache optimization

meshopt_optimizeVertexCache(indices.data(), indices.data(), index_count, vertex_count);

Reorders triangles for GPU post-transform vertex cache. Reduces vertex shader invocations.

Pass 2 — Overdraw reduction

meshopt_optimizeOverdraw(indices.data(), indices.data(), index_count,
                          vertex_positions, vertex_count, vertex_stride, 1.05f);

Clusters triangles by depth to reduce pixel shader overdraw.

Pass 3 — Vertex fetch optimization

meshopt_optimizeVertexFetch(vertices.data(), indices.data(), index_count,
                             vertices.data(), vertex_count, vertex_stride);

Reorders vertices to match the optimized index order — reduces cache misses on vertex fetch.

Expected impact

Scene	Before	After (estimated)
Bistro exterior (.zemesh)	292 MB	~35–50 MB
Render frame time	Baseline	~15–30% improvement
GPU vertex cache hit rate	~40–60%	~85–95%

Integration

• Add via FetchContent_Declare in dependencies.cmake (single source file pattern like ufbx)
• Call after vertex dedup in FbxImporter::ImportFile and GltfImporter::ImportFile
• Gate behind ImportOptions.OptimizeMesh = true (default on) in the UI

Prerequisite

Fixes the cold-start crash from large .zemesh files (#668 follow-on).