Provide functionality of interpolating real and imaginary parts and calculate the tridiagonal covariance matrix all at once

[BjoernLudwigPTB]

For some cases of complex interpolations we need this functionality to avoid (unwanted or unthought-of) side effects of zero padding or similar methods. As a first step we prepare now for leaving the covariance calculation to the user by providing the sensitivity coefficients (see #111), but we could as well go the full distance and provide the covariance matrix as well.

In this case we might have to reformulate the uncertainty equations to take care of the correlation between timestamps and real and imaginary parts. We could tackle that by writing the according linear equations in (block) matrix form.

[BjoernLudwigPTB]

@eichstaedtPTB Your suggestion was tackle this by wrapping the according calls of interp1d_unc() in propagateDFT() where this will be needed, right?