DIFFA-2: A Practical Diffusion Large Language Model for General Audio Understanding | AI Papers Podcast Daily | Podwise