Optimize the performance of tensor.split.map

[Atry]

We should create a view for the Seq returned by tensor.split.map, instead of actually creating Tensors for each element in the Seq