Special handling for op_cat for 1D empty tensor

Summary:
The behavior is a bit different. When we see a 1D empty tensor in the tensor list, just ignore it and treat it as a wildcard, and go to the next one.  The dim rules doesn't apply here.

From aten:  (can use `torch.ops.aten.cat.out` instead of `torch.cat`)
```
>>> torch.cat([torch.ones([0]), torch.ones([0])], dim=-4)
tensor([])
>>> torch.cat([torch.ones([0]), torch.ones([0]), torch.ones([1, 1])], dim=0)
tensor([[1.]])
>>> torch.cat([torch.ones([0,0,0,0]), torch.ones([0,0,0])], dim=2)
Tensors must have same number of dimensions: got 4 and 3
>>> torch.cat([torch.ones([0]), torch.ones([0]), torch.ones([0]), torch.ones([1])], dim=1000)
Dimension out of range (expected to be in range of [-1, 0], but got 1000)
>>> torch.cat([torch.ones([0]), torch.ones([0,0])], dim=-4)
Dimension out of range (expected to be in range of [-2, 1], but got -4)

Reviewed By: manuelcandales

Differential Revision: D48449555

fbshipit-source-id: 3eb4559b7229cddf78e2c70adacf83969d454643