Do not test that the returned values are close to the expected values (it is not clear which number is first and which is second). Suppose that the user function returns 2 complex numbers a and b. Then the tests should check that a + b is close to sum and a * b is close to product. I also recommend to use relative errors when checking that two floating-point numbers are close.

Do not ask to return a dynamically allocated array with a fixed number of elements. It is better to return a struct with 4 fields (but for this task it is much better to use complex numbers directly):

struct answer {
  double complex a;
  double complex b;
};