Cannot convert the endoscopic_tool_segmentation pretrained model to torchscrpit

[binliunls]

Describe the bug
When running the torchscript export command line of endoscopic_tool_segmentation bundle in MONAI model zoo, got the error below:

Missing key(s) in state_dict: "decoder.blocks.0.upsample.preconv.weight", "decoder.blocks.0.upsample.preconv.bias", "decoder.blocks.1.upsample.preconv.weight", "decoder.blocks.1.upsample.preconv.bias", "decoder.blocks.2.upsample.preconv.weight", "decoder.blocks.2.upsample.preconv.bias", "decoder.blocks.3.upsample.preconv.weight", "decoder.blocks.3.upsample.preconv.bias".                                 
        size mismatch for decoder.blocks.0.convs.conv_0.conv.weight: copying a param with shape torch.Size([256, 472, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 296, 3, 3]).                                                                       
        size mismatch for decoder.blocks.1.convs.conv_0.conv.weight: copying a param with shape torch.Size([128, 304, 3, 3]) from checkpoint, the shape in current model is torch.Size([128, 176, 3, 3]).                                                        
        size mismatch for decoder.blocks.2.convs.conv_0.conv.weight: copying a param with shape torch.Size([64, 152, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 88, 3, 3]).                                                            
        size mismatch for decoder.blocks.3.convs.conv_0.conv.weight: copying a param with shape torch.Size([32, 80, 3, 3]) from checkpoint
, the shape in current model is torch.Size([32, 48, 3, 3]).    

This is mainly caused by the PR #5982, since it just changed the pre_conv parameter from None to default. And the previous model is trained under None mode.

After added this pre_conv parameter to the FlexibleUNet and set it to None in the bundle, still got a type hint error:

Expression of type | cannot be used in a type expression:
File "/home/liubin/data/github_monai/export_trt_build/MONAI/monai/networks/nets/basic_unet.py", line 152
    def forward(self, x: torch.Tensor, x_e: torch.Tensor | None):
                                 ~~~~~~~~~~~~~~~~~~~ <--- HERE

This is caused by the torch.jit.script cannot convert the | type in the type hints.

To Reproduce
Steps to reproduce the behavior:

1. Start a monai docker
2. Pull the model-zoo repo
3. Run the ckpt_export commandline in the endoscopic_tool_segmentation bundle

Expected behavior
Successfully converted the model to torchscript.

Screenshots
If applicable, add screenshots to help explain your problem.