update and improve error_budget_pie function

[peterstangl]

Previously, the error_budget_pie function was representing the individual errors as fractions of a linearly summed total error. Such a representation can be rather misleading since the actual total error corresponds to the individual errors summed in quadrature, not linearly.

This PR updates the error_budget_pie function to improve the visual representation of the individual error contributions. In the new version

• the wedges are labelled with the actual relative error contributions (not as the fractions of the linearly summed total error as before),
• the area of each wedge is proportional to the squared error contribution such that the total area of the pie corresponds to the actual squared total error.

So the new version gives an accurate graphical representation of the individual errors summed in quadrature and the sizes of the wedges have an intuitive interpretation in terms of the individual variance contributions that can be summed linearly. Furthermore, the numbers on the labels now exactly correspond to the results of flavio.sm_error_budget, also when small contributions are grouped together under "other" (this was not the case before since "other" always corresponded to small errors summed in quadrature), and summing all the numbers on the labels in quadrature now exactly yields the actual total error.

[DavidMStraub]

Why not np.array(err_dict.values())? Perhaps even sum(v**2 for v in err_dict.values())?

[peterstangl]

err_dict.values() is a dict_values object and np.array(err_dict.values()) is an array with dtype=object containing a single element. So for using numpy, one needs to convert the dict_values into a list first. But it's true that in this case, maybe using numpy is not the best option and the simple sum(v**2 for v in err_dict.values()) might be more clear. And it is actually faster than the numpy version.