Free GPU memory between HunyuanFoley segments to prevent OOM

After each segment's denoise_process, explicitly del audio_batch and
visual_feats then call torch.cuda.empty_cache(). The 15-s audio latent
tensor is several GB; without explicit deletion PyTorch holds the CUDA
allocation until GC runs, causing OOM when the second segment allocates
its own latent. This is why seg 1 completed successfully but seg 2 failed
silently (ZeroGPU kills worker on OOM with no Python traceback).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app.py

@@ -1301,6 +1301,12 @@ def _hunyuan_gpu_infer(video_file, prompt, negative_prompt, seed_val,
                     batch_size=1,
                 )
                 seg_wavs.append(audio_batch[0].float().cpu().numpy())
+                # Free GPU memory between segments — latents/visual_feats from denoise_process
+                # stay allocated until GC runs; explicit deletion + cache clear prevents OOM
+                # when processing a second segment (the 15-s latent tensor is ~several GB).
+                del audio_batch, visual_feats
+                if torch.cuda.is_available():
+                    torch.cuda.empty_cache()
 
             _log_inference_timing("HunyuanFoley", time.perf_counter() - _t0,
                                   len(segments), int(num_steps), HUNYUAN_SECS_PER_STEP)