Improve BPM fix by interpolating in the dispersion axis

[sergiopasra]

By default BPM correction is computed as the median of valid pixels in a 5x5 box. In MEGARA, this approach mixes bright pixels from the peak of the fibers with dark values from the valley between fibers. In my tests, I get better results if the box contains values only in the X direction (like 7x1)