Avoid integer overflow in EigenMetaKernel indexing

- The current implementation computes `size + total_threads`, which can
  overflow and cause CUDA_ERROR_ILLEGAL_ADDRESS when size is close to
  the maximum representable value.
- The num_blocks calculation can also overflow due to the implementation
  of divup().
- This patch prevents these overflows and allows the kernel to work
  correctly for the full representable range of tensor sizes.
- Also adds relevant tests.

unsupported/Eigen/CXX11/src/Tensor/TensorMeta.h

@@ -30,13 +30,15 @@ const T2& choose(Cond<false>, const T1&, const T2& second) {
 template <typename T, typename X, typename Y>
 EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
 T divup(const X x, const Y y) {
-  return static_cast<T>((x + y - 1) / y);
+  // Note: This form is used because it cannot overflow.
+  return static_cast<T>(x == 0 ? 0 : (x - 1) / y + 1);
 }
 
 template <typename T>
 EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
 T divup(const T x, const T y) {
-  return static_cast<T>((x + y - 1) / y);
+  // Note: This form is used because it cannot overflow.
+  return static_cast<T>(x == 0 ? 0 : (x - 1) / y + 1);
 }
 
 template <size_t n> struct max_n_1 {