Generic fitting model

[mpizzocaro]

GaussMarkovSRS handles inputs that are absolute frequency values or frequency ratios.
This means that the model for input data is either $q_j$ or $q_i/q_j$, where the $q_j$ are the adjusted constants (and in the current implementation, Cs is handled as a special case).
It may be worthwhile to rework the code and the input data format do handle generic models $\prod_j {q_j}^{n_j}$.
For example, this would enable GaussMarkovSRS to handle adjustment in the style of CODATA.