[Export] Introduce as_none in ex.Argument union type (#93210)
This design has two implications
- We are **NOT** modeling nullable argument types, e.g `Tesnor?`, `int?`, `int[]?` as a special argument type
- Python None is treated as a special argument type, downstream executor/runtime need know to handle this.
For aten.convolution's schmea, it accepts an optional input: `Tensor? bias`
```
convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor
```
Example: notice the **None** argument in the following fx.node
```
convolution_default = torch.ops.aten.convolution.default(arg0, _param_constant0, None, [2, 2], [3, 3], [1, 1], False, [0, 0], 1)
```
would be exported as
```
Node(
op='call_function',
target='aten.convolution.default',
args=[
Argument(as_tensor=TensorArgument(name='arg0')),
Argument(
as_tensor=TensorArgument(name='_param_constant0')
),
Argument(as_none=True),
Argument(as_ints=[2, 2]),
Argument(as_ints=[3, 3]),
Argument(as_ints=[1, 1]),
Argument(as_bool=False),
Argument(as_ints=[0, 0]),
Argument(as_int=1)
],
kwargs={},
outputs=[
ReturnArgument(
as_tensor=TensorArgument(name='convolution_default')
)
],
metadata='Skipped'
),
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/93210
Approved by: https://github.com/suo
