GitLab

Fix tensor numel() return dtype

The original dytpe int would be an issue if the tensor element numbers are larger than int32's range (which is not a super large number in real workloads)
Signed-off-by: Shang Zhang <shangz@nvidia.com>