Make core modules TorchScript compatible

[MarcelloCeresini]

Summary

This PR updates exportPythonModel so that the generated TracerModel classes can be scripted with torch.jit.script.

What Changed

nnodely/exporter/export.py

• Generated Python model files now import Dict and Tuple
• TracerModel.forward now has a TorchScript compatible type signature
• Output dictionaries are now assigned with torch.jit.annotate(Dict[str, torch.Tensor], ...) so their types are visible and constant for TorchScript
• The generated update now uses safe non-mutable defaults None instead of {}, and is also marked with torch.jit.ignore since it's not needed for inference

tests/test_export_native_torchscript.py

• Builds a representative nnodely model with standard classes
• Exports it with exportPythonModel
• Imports the generated Python module
• Runs eager backward
• Run torch.jit.script(TracerModel())
• Run scripted backward
• Save and reload the scripted model
• Verify eager vs scripted vs reloaded outputs match

Why

TorchScript exports make nnodely models easier to deploy outside Python, in particular they can be loaded from C++ applications through LibTorch, enabling integration into simulators or embedded systems with a small change to the generated export code.