Update to hspline and lowess filters.

[talensgj]

Hello,

I've made some modifications to the hspline and lowess filters, specifically:

hspline

• Add StandardScaler to improve stability.
• Call HuberRegressor with max_iter=1000 to improve stability/convergence.
• Expose HuberRegressor epsilon through cval.

lowess

• Change the way the left and right indices of the window are computed to use np.searchsorted, fixing a bug.
• Simplify calc_x_weights, by changing y_weights intilisation.
• Expose the y_weight cut-off through cval.

I hope you will find these changes make sense and merge them into wotan.

[hippke]

Hi, Thank you for your proposed changes! For each of them I have some questions. Let's start with the easiest ones:
Why change cval = 1.35 to cval = 3 in hspline, i.e. why choose this specific value? The default numbers were chosen in alignment with the reference papers.

[talensgj]

@hippke I used cval=3 in my work, because in my tests cval=1.35 seemed a bit aggressive. The current state of the PR still has the value at cval=1.35, I have merely made it so the user can change it if they want.