repr of integer tensors with large values doesn't match construction

At least in python 3 (I didn't check python 2), literals in scientific notation are always floats.

So:
```
>>> a=torch.tensor(425245252)
>>> a.dtype
torch.int64
>>> a
tensor(4.2525e+08)
>>> torch.tensor(4.2525e+08).dtype
torch.float32
```

CC @li-roy 