Add pivot point to power law model to reduce parameter correlations

[pkienzle]

Playing with power law and DREAM, I see a strong negative correlation between scale and power. This correlation can be removed by defining a pivot point in the center of the data so that power rotates the log-log curve around this point and scale shifts it up and down:

I(q) = scale * (q/pivot)**(-power) + background

If you pick the pivot in the middle of the q range then fit scale and power, the result is a nicely independent correlation plot with no change in the uncertainty of the fitted power. The correlations get stronger the further the pivot is from the center.

Default pivot to one for the existing behaviour. The fitted scale value should be near the measured I(pivot).

Similar parameters from other models:

peak_pos: broad_peak, gaussian_peak, peak_lorentz
q_0: spinodal