Try to put the list of output formats for vf_scale in a more sensible order. While the 16-bit yuv formats should not be preferred over the 8-bit ones, it seems reasonable to probe them directly after the equivalent 8-bit formats.