Merge input and output shape when removing identity (#2588)

Similar with #2578, for this case:

```python
import torch
import torch.nn as nn


class Model(nn.Module):
    def forward(self, x):
        return x.new_zeros(x.shape)


def main():
    model = Model()
    args = torch.rand(4, 4),
    batch = torch.export.Dim("batch")
    dynamic_shapes = {"x": {0: batch}}
    torch.onnx.export(
        model,
        args,
        "model_test.onnx",
        dynamic_shapes=dynamic_shapes,
        dynamo=True,
    )


if __name__ == "__main__":
    main()
```

---------

Co-authored-by: Justin Chu <[email protected]>

Author: wodesuck

onnxscript/optimizer/_constant_folding.py

@@ -608,6 +608,9 @@ def identity(node: ir.Node, op, state: OptimizerState) -> ReturnValue:
     input = node.inputs[0]
     output = node.outputs[0]
     if input is not None and output is not None:
+        input.shape = _merge_shapes(input.shape, output.shape)
+        if input.type is None:
+            input.type = output.type
         state.set_sym_value(output, input)
     return None