don't delegate nodes with empty tensors (#1847)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/1847

we occaisionally see the following in some graphs:
```
%empty_20 : [num_users=1] = call_function[target=torch.ops.aten.empty.memory_format](args = ([0],), kwargs = {device: cpu, pin_memory: False})
%cat_97 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%slice_146, %slice_145, %empty_20],), kwargs = {})
```

In which empty_20 is an empty tensor. xnnpack does not know how to deal with empty tensors or [0] shape tensors. So don't delegate.

Reviewed By: digantdesai

Differential Revision: D53398027

fbshipit-source-id: 3f8b113b059a9ab04890a756b1bc8557b54f8fce