Dynam3D: Dynamic Layered 3D Tokens Empower VLM for Vision-and-Language Navigation | Read Paper on Bytez