Weighted Tailfit

[hschnait]

Dear all

This PR implements the weighted tailfit originally proposed by @gkraberger (PR #482), that got lost during the rework of the tails in 2.x (see also #719).

Please review whether you want to add it in this way - or if I'm messing with something I should not mess with here.
The default behaviour of the tailfit does not change, the order of the error can be set by set_tail_fit_parameters.

[the-hampel]

@Wentzell can we pick this up again and maybe merge? I thing this might be a very useful feature, because the default tail fit by fraction is rather useless at the moment without weighting by frequencies, as the tail fit is dominated by Gf elements from frequencies with very larger error. I checked a bit the commits, and the changes seem to be rather small and do not break any backward compatibility if I see this correctly.

[Wentzell]

Yes we should definitely pick this up again!

The current implementation is a bit stringent as it limits you to a particular analytic form for the fitting weights.
We should generalize and have the user provide a function that returns the weight based on Matsubara index.