Energy-Structured Latent World Models with Neural Time Fields for Physically Constistent Open-World Motion Planning