Scaling Diffusion Language Models via Adaptation from Autoregressive Models