Stopped ProxyTensor from turning aten::lift tensors into proxy objects (#81024)
```
def f():
val = torch.tensor(float('inf'))
return torch.full((100, 100), val)
```
today we turn `val` into a ProxyTensor, and then complain when we try to turn `val` into a scalar.
We call `aten::lift` when we call `torch.tensor(5)`, so this just prevents those from being turned into ProxyTensors unnecessarily.
cc: @ezyang
Pull Request resolved: https://github.com/pytorch/pytorch/pull/81024
Approved by: https://github.com/ezyang
