Update modulation noise to match performance model's current best estimate

We currently use a fit to Simon Barke's thesis (figure 5.13, blue curve). This curve corresponds to a 1-W fiber amplifier, which is not realistic.

We should use a fit to the red curve, which is what the Performance Model (dec. 2020) uses (section 12.3.1). In fractional frequency deviations (to be multiplied by the modulation frequency), it reads 5.2 \times 10^{-14} f^{1/3}.