linalg::Sum(const Tensor&) gives wrong answer for complex types

[IvanaGyro]

The test below should pass.

#include "cuda_runtime_api.h"
#include "gtest/gtest.h"

#include "cytnx_error.hpp"
#include "linalg.hpp"
#include "Tensor.hpp"
#include "Type.hpp"

using namespace cytnx;

TEST(Linalg, Sum) {
  Tensor block(/* shape */ {5}, Type.ComplexDouble, Device.cuda, /* init_zero */ true);
  checkCudaErrors(cudaSetDevice(block.device()));
  block.fill(cytnx_complex128{2., 3.});
  Tensor sum_result = cytnx::linalg::Sum(block);
  checkCudaErrors(cudaDeviceSynchronize());

  EXPECT_EQ(sum_result.shape().size(), 1);
  EXPECT_EQ(sum_result.shape()[0], 1);
  EXPECT_DOUBLE_EQ(sum_result.at<cytnx_complex128>({0}).real(), 10.);
  EXPECT_DOUBLE_EQ(sum_result.at<cytnx_complex128>({0}).imag(), 15.);
}

However, the test fails.

[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from Linalg
[ RUN      ] Linalg.Sum
test.cpp:20: Failure
Expected equality of these values:
  sum_result.at<cytnx_complex128>({0}).real()
    Which is: 0
  10.
    Which is: 10

test.cpp:21: Failure
Expected equality of these values:
  sum_result.at<cytnx_complex128>({0}).imag()
    Which is: 0
  15.
    Which is: 15

[  FAILED  ] Linalg.Sum (496 ms)
[----------] 1 test from Linalg (496 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test suite ran. (496 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] Linalg.Sum

 1 FAILED TEST


0% tests passed, 1 tests failed out of 1

Total Test time (real) =   4.98 sec

This issue may be the culprit for failing the tests BlockUniTensorTest.gpu_Trace and DenseUniTensorTest.gpu_Trace.