Feature: Smoothing fine-grained calibration measurements

[dekuenstle]

Measuring fine-grained steps in the calibration routine reveal noise from the measurement device and display, which results in a non-monotonic luminance function. The current linear-interpolation method cannot invert this function without dropping the non-monotonic steps (order of magnitude: ~0 steps of 200 measurements, but ~1000 steps of 4000 measurements).
An improved calibration routine should smooth the measurements e.g. by applying a filter or a non-linear statistical model.
Additionally, this model could be useful for measuring the noise itself.

A fast solution could apply a simple filter, e.g. a uniform filter (https://stackoverflow.com/a/73991730) or a Savitzky-Golay filter.
A more advanced solution could be Locally-linear models (LOESS) could be a good start, as they are fast, relatively reliable (https://stackoverflow.com/a/63458548), and available in the statmodels package.