Review convolution with resolution function

Convolution of measured signal with resolution function is implemented in Device/Resolution/Convolve.cpp. Foundations are described in paper by Jeremy Fix, ba-intern/lit/JFix_FFTConvolution.pdf.

The current code supports several modes, of which only one FFTW_LINEAR_SAME is actually used. The others can probably be removed.

Our resolution functions are symmetric. This should result in a substantial simplification. Possibly, we can get rid of all imaginary parts.

Hopefully, we can get rid of negative output in example RoughAndSpecular (#864).