expected modeling could be sped up

[Thomas Gilgenast]

expected modeling currently calls statsmodels.nonparametric.smoothers_lowess.lowess(), but if it called lib5c.util.lowess.lowess_fit() or used the delta kwarg it would be a lot faster

the donut filters currently call scipy.ndimage.generic_filter() but if they used scipy.signal.convolve() they would be much faster as well

[Thomas Gilgenast]

fixes #35

untested commit

→ <<cset 244428cf52b5>>