Add supervised torchfort_save_model/torchfort_load_model API tests.

[romerojosh]

Summary

Adds supervised test coverage for the public torchfort_save_model / torchfort_load_model API.

The new coverage exercises both native MLP models and TorchScript models, including loading saved model weights across device combinations. Each case verifies that:

• inference matches after loading saved model weights into a fresh model
• the loaded model remains trainable on the destination device

Changes

• Added a trainable TorchScript test fixture generated by tests/supervised/scripts/setup_tests.py
• Added configs/torchscript_trainable.yaml
• Installed the new config through the supervised test CMake rules
• Added model-only save/load coverage in test_checkpoint for:
  • native MLP
  • TorchScript
  • CPU -> CPU
  • GPU -> GPU
  • CPU -> GPU
  • GPU -> CPU
  • GPU0 -> GPU1 when 2 GPUs are available

[romerojosh]

/build_and_test

[github-actions]

🚀 Build workflow triggered! View run

[github-actions]

❌ Build workflow failed! View run

[romerojosh]

/build_and_test

[github-actions]

🚀 Build workflow triggered! View run

[github-actions]

✅ Build workflow passed! View run

[azrael417]

LGTM, all tests make sense.