ONNX Conversion of DS1 graph fails

[bsarden]

Currently, the DS1 graph fails to export to ONNX due to the HardTanh operation not being part of the supported ONNX operators.

torch.onnx.export(model._get_network(), dummy_input, model_path, verbose=True, input_names=input_names)

> RuntimeError: ONNX export failed: Couldn't export operator aten::hardtanh

A simple fix could be to replace the nn.HardTanh operator with a custom layer that just uses torch.clamp instead, but there might be some additional code associated with module.layers, since torch.clamp expects a Tensor (i.e. not a Module). Hopefully, torch.onnx.export is smart enough convert a torch.clamp() call into the appropriate Clip operation(s).