Tensor numel() return dtype to be size_t (#2022)

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>

transformer_engine/common/common.h

@@ -88,7 +88,7 @@ struct SimpleTensor {
             nvte_make_shape(this->shape.data(), this->shape.size())};
   }
 
-  int numel() const {
+  size_t numel() const {
     size_t acc = 1;
     for (const auto &dim : shape) {
       acc *= dim;