Use torchvision weights API in scriptmodule scripts

[vishesh9131]

Description of the change

The three scriptmodule export scripts load pretrained models with pretrained=True, which torchvision has deprecated since 0.13 and warns about on every run:

UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.

This switches them to the weights= API:

• create_resnet18_scriptmodule.py: ResNet18_Weights.IMAGENET1K_V1
• create_encoder_cnn_backbone_scriptmodule.py: ResNet50_Weights.IMAGENET1K_V1
• create_vgg19_layers_scriptmodule.py: VGG19_Weights.IMAGENET1K_V1

IMAGENET1K_V1 is what pretrained=True loaded for all three, so the exported files keep the same weights. For ResNet-50 this matters: ResNet50_Weights.DEFAULT is IMAGENET1K_V2, a different checkpoint, so the weights are pinned explicitly (with a short comment) instead of using DEFAULT.

The weights= API was added in torchvision 0.13, which pairs with PyTorch 1.12, the minimum version the README supports.

Testing

I regenerated each file with the old and the new script (PyTorch 2.10, torchvision 0.25) and compared every tensor with torch.equal:

Script	Tensors compared	Result
create_encoder_cnn_backbone_scriptmodule.py	319	identical
create_vgg19_layers_scriptmodule.py	32	identical
create_resnet18_scriptmodule.py	80 learned parameters (randomly initialized fc head excluded)	identical

For ResNet-18, the 40 BatchNorm running_mean/running_var buffers differ between old and new, but they also differ between two runs of the unchanged script: it traces the model without calling model.eval(), so the forward pass during tracing updates them with the random example input. That is pre-existing and independent of this change; I'm happy to send a separate fix for it.

The new scripts print no deprecation warnings (the old ones print two per run), and the C++ tutorials that load these files (pytorch-basics, neural-style-transfer, image-captioning) run with the regenerated files.

Type Of Change

• Bug Fix (non-breaking change that fixes an issue)
• New Feature
• New PyTorch tutorial
• Breaking Change (cmake changes, fix or feature that would cause existing functionality to not work as expected)

Related Issues

None; found while running all tutorials locally (no existing issue).

Development & Code Review

• cpplint rules passes locally (run cmake -P cpplint.cmake)
• CI is passing
• Changes have been reviewed by at least one of the maintainers