Include resolution function uncertainties by propagation through convolution

[andyfaff]

The measurement uncertainties used in the fitting examples only use those of the measured QENS spectrum:

    Mq = bmp.Curve(model_convol, x, data, error, q=q_5A[i],
               scale=20, center=0.0, A0=0.0, A1=0.9, hwhm1=0.05, hwhm2=0.3,
               resolution=resol)

Surely they should be including the uncertainties of the measured resolution function as well? The uncertainties package would probably allow for the resolution uncertainties to be propagated through a convolution with the model. However, I'm not sure of the correct formulation for calculating a log-likelihood against the measured data at that point.