Add TensorViewBuilder::shape(std::vector<Val*> shape)

[zasdfgbnm]

This will be needed to build out the arange support with symbolic start, end, and step, because the size of its output is ceilDiv(sub(end, start), step)

[naoyam]

The changes look good to me, but can you elaborate a little bit about why it's needed for arange?

[zasdfgbnm]

For arange, I would need something like this:

TORCH_CUDA_CU_API TensorView* arange(Val* start, Val* end, Val* step, DataType dtype) {
  auto size = castOp(DataType::Int, ceilDiv(sub(end, start), step));
  auto out = TensorViewBuilder().ndims(1).dtype(dtype).contiguity({true}).shape({size}).build();
  IrBuilder::create<ARangeOp>(out, start, end, step);
  return out;
}

[naoyam]

LGTM