Implement orthogonal polynomials (or better yet bsplines) for detrending

[gdevenyi]

Just squaring the timeseries can lead to numerical instability, and isn't as flexible as a spline basis for the timeseries detrending. At least use orthogonal polynomials here.

Ideally, switch (or add) a a bspline detrending and allow the user to specify the order (1=linear, 2+ "higher order curvyness")