[PyTorch] Don't use scientific notation for printing integer tensors

We currently use scientific notation (E-notation) to print large integer tensors. We shouldn't do this because it hides meaningful digits and because E-notation corresponds to floating point data types in Python.

```python
>>> x = torch.ones(525076, dtype=torch.int)
>>> x.sum()
tensor(5.2508e+05)
>>> x.sum().item()
525076
```