Currently scaler_np (the core function in AdaptiveStandardScaler) requires that the target axis has a .gain field.
To support CoordinateAxis, without gain, we could calculate the alpha for each sample or chunk (using average gain). For this to work, the EWMA class would have to be refactored so the kernel ([alpha], [1.0, alpha - 1.0]) can be recalculated without resetting internal state.