force contiguous when calling `augment_model_with_bundled_inputs`

Summary: `augment_model_with_bundled_inputs` can compress the tensor when values are constant, however it requires contiguous layout and `zero_like` can return non-contiguous ones

Reviewed By: zhanghang1989

Differential Revision: D28224987

fbshipit-source-id: 32b13728ff8fadd53412dbf2d59c4b46e92af04a

d2go/export/torchscript.py

@@ -85,7 +85,7 @@ def trace_and_save_torchscript(
             iters = recursive_iterate(inputs)
             for x in iters:
                 if isinstance(x, torch.Tensor):
-                    iters.send(torch.zeros_like(x))
+                    iters.send(torch.zeros_like(x).contiguous())
             inputs = iters.value
             augment_model_with_bundled_inputs(liteopt_model, [inputs])
             liteopt_model.run_on_bundled_input(0)  # sanity check