Fix generate_diffusion_uncond: remove unsupported mask arg in sample_k call

[kmraven]

This PR fixes a runtime error in generate_diffusion_uncond caused by passing an unexpected mask argument to sample_k.

In stable_audio_tools/inference/generation.py, line ~78, the generate_diffusion_uncond calls:

sampled = sample_k(model.model, noise, init_audio, mask, steps, **sampler_kwargs, device=device)

However, sample_k (defined in stable_audio_tools/inference/sampling.py) expects the 4th positional argument to be steps and does not accept a mask parameter at all. Other call sites use sample_k without mask.