Cannot correctly compute cross entropy loss (backward) when the reduction used is "sum" and torch executor is used.

Nobody has claimed this yet.

*Note*: If you have a model or program that is not supported yet but should be, please use the program coverage template.

## 🐛 Bug

With the torch executor, Thunder cannot correctly compute cross entropy loss (backward) when the reduction used is "sum". It works correctly with "mean"

### To Reproduce

Run the code snippet below.

#### Code sample

```
device='cuda'
executor=TorchExecutor

def cross_entropy_fn(logits, labels):
return torch.nn.functional.cross_entropy(logits, labels, reduction="sum")

sequence_length, vocab_size = 8192, 32064
logits = torch.rand((sequence_length, vocab_size), device=device, dtype=thunder_dtype, requires_grad=True)
labels = torch.randint(0, sequence_length, (sequence_length,), requires_grad=False, device=device)

fn = executor.make_callable(cross_entropy_fn)
a = fn(logits, labels)
a.backward()
print(logits.grad)
```

### Expected behavior

Should run correctly to completion with output matching that of torch compile.

### Environment

- PyTorch Version (e.g., 1.0):
- OS (e.g., Linux):
- How you installed PyTorch (`conda`, `pip`, source):
- Build command you used (if compiling from source):
- Python version:
- CUDA/cuDNN version:
- GPU models and configuration:
- Any other relevant information:

### Additional context

The error seen is:
https://gist.github.com/protonu/ff88fbc9508947f45e39a6190acc6508

Start with the cross_entropy_fn reproducer in this issue and inspect the linked error gist, comparing the TorchExecutor behavior for reduction="sum" with reduction="mean" and torch.compile. The work is done when the backward pass completes and its output matches torch.compile for the provided case.

Written by the indexing model from the issue text.