Fix complex math abs and abs2 issues

[sbastrakov]

The issues were introduced in #3245 and went unnoriced until #3379 .

Fix wrong formula in Abs implementation.
Make return types of Abs and Abs2 real, as they were before that change.

Fixes #3379 .

[psychocoderHPC]

@BeyondEspresso PLease check if this solves you issues

[BeyondEspresso]

For documentation, the PR created following compile error

/home/debus/src/picongpu/thirdParty/cupla/alpaka/include/alpaka/math/sqrt/Traits.hpp(51): error #70: incomplete type is not allowed
          detected during:
            instantiation of "auto alpaka::math::sqrt(const T &, const TArg &) [with T=alpaka::math::AbsUniformCudaHipBuiltIn, TArg=picongpu::float_X]" 
/home/debus/src/picongpu/include/pmacc/../pmacc/math/complex/Complex.tpp(224): here
            instantiation of "auto alpaka::math::traits::Abs<T_Ctx, pmacc::math::Complex<T_Type>, void>::abs(const T_Ctx &, const pmacc::math::Complex<T_Type> &)->T_Type [with T_Ctx=alpaka::math::AbsUniformCudaHipBuiltIn, T_Type=picongpu::float_X]" 
/home/debus/src/picongpu/thirdParty/cupla/include/cupla/device/math/Abs.hpp(37): here
            instantiation of "auto cupla::cupla_cuda_async::device::math::abs(const T_Type &)->decltype((<expression>)) [with T_Type=pmacc::math::Complex<picongpu::float_X>]" 
/home/debus/src/picongpu/include/pmacc/../pmacc/math/complex/Complex.tpp(182): here
            instantiation of "auto alpaka::math::traits::Sqrt<T_Ctx, pmacc::math::Complex<T_Type>, void>::sqrt(const T_Ctx &, const pmacc::math::Complex<T_Type> &)->pmacc::math::Complex<T_Type> [with T_Ctx=alpaka::math::SqrtUniformCudaHipBuiltIn, T_Type=picongpu::float_X]" 
/home/debus/src/picongpu/thirdParty/cupla/include/cupla/device/math/Root.hpp(37): here
            instantiation of "auto cupla::cupla_cuda_async::device::math::sqrt(const T_Type &)->decltype((<expression>)) [with T_Type=pmacc::math::Complex<picongpu::float_X>]" 
/home/debus/src/picongpu/include/pmacc/../picongpu/fields/background/templates/TWTS/EField.tpp(334): here

[sbastrakov]

So I made a mistake in the fixed Abs implementation. Not sure what is it from the first look, will investigate tomorrow morning. Thanks for testing @BeyondEspresso , sorry it did not work as i imagined it would.

[sbastrakov]

@BeyondEspresso there was a tricky error in my original implementation. It is hopefully fixed (at least my small test compiled, but of course you have a more complex use case) now with the current state of this branch

[BeyondEspresso]

@sbastrakov I ran my test case with your new branch and the results of the old master and your PR are virtually the same: