Convert BatchedDGL graph to cuda tensor

How do I convert a dgl.batched_graph.BatchedDGLGraph object to cuda?
Just doing .to(device) gives a NoneType object

to operation is different from PyTorch, it’s an inplace operation. g.to(device) will return nothing but move g to new device.

Was there a reason to do so? Would be great if consistent with PyTorch

This PR should fix the problem.