Use randn_tensor to replace torch.randn (huggingface#10535)

`torch.randn` requires `generator` and `latents` on the same device, while the wrapped function `randn_tensor` does not have this issue.

Author: lmxyy

src/diffusers/pipelines/ltx/pipeline_ltx.py

@@ -769,7 +769,7 @@ def __call__(
             if not self.vae.config.timestep_conditioning:
                 timestep = None
             else:
-                noise = torch.randn(latents.shape, generator=generator, device=device, dtype=latents.dtype)
+                noise = randn_tensor(latents.shape, generator=generator, device=device, dtype=latents.dtype)
                 if not isinstance(decode_timestep, list):
                     decode_timestep = [decode_timestep] * batch_size
                 if decode_noise_scale is None: