Windowed filters get a wrong window
The get_window() method does not behave properly: if we have a data set (like the IMIS COR2 around 2014-01-09T12:30) that contains a big gap just before the points we try to filter (with MAD), both the nr_points AND time criteria will be fulfilled because it will use the points from 2 days ago to fill the window, which is a non-sense in term of filtering.