Harden mergeMeshes by preparing candidates and improving diagnostics

[tracygardner]

Motivation

• mergeMeshes should normalize candidate meshes the same way intersectMeshes does so composite/child-only meshes and non-indexed geometry are handled reliably.
• Provide clear, per-mesh diagnostics when CSG conversions fail to make multi-merge failures actionable.
• Avoid running costly or fragile CSG loops when only zero or one valid mesh remains.

Description

• Updated flockCSG.mergeMeshes(modelId, meshList) to run each candidate through prepareMeshForCSG before further checks and skipped meshes that return null with a mesh-name warning via console.warn.
• Short-circuited prepared-mesh count: return null when none remain, and return/clone the single prepared mesh when exactly one remains (preserving metadata and blockKey) without entering the CSG add loop.
• Wrapped each flock.BABYLON.CSG2.FromMesh(...) call (base mesh and each additional mesh) in targeted try/catch blocks that log which mesh failed conversion before re-throwing to trigger fallback cleanup.
• Preserved the fallback BABYLON.Mesh.MergeMeshes path and ensured it only receives prepared meshes.
• Added integration tests in tests/materials.test.js covering: single simple mesh merge, multiple valid meshes merge, one valid + one composite/child-only merge, and one valid + one invalid mesh (asserts warning and a valid result). Files changed: api/csg.js, tests/materials.test.js.

Testing

• Added new integration coverage in tests/materials.test.js for the four scenarios described above (tests present but not fully executed here).
• Performed syntax checks with node --check api/csg.js and node --check tests/materials.test.js which succeeded.
• Attempted to run npm run test:api materials but it failed in this environment because Playwright’s Chromium browser binary is not installed (npx playwright install required), so browser-based tests could not be executed here.

---

Codex Task