Fix Low GPU Utilization

https://github.com/GuyTevet/motion-diffusion-model/blob/db5478ab1ba4592607b54f1c25542f24b4f6a60f/diffusion/gaussian_diffusion.py#L1321

causes low gpu utilization. Using it as a context manager as intended solves this problem.

```python
with torch.autograd.set_detect_anomaly(True):
```

Start at diffusion/gaussian_diffusion.py around line 1321 and inspect how torch.autograd.set_detect_anomaly is used. Confirm the current usage is causing low GPU utilization and compare it with the context-manager form shown in the issue. Done means anomaly detection uses the intended context-manager form without changing unrelated behavior.

Written by the indexing model from the issue text.