videoconvert: disregards downstream preferences for the format

With such a pipeline:

src ! videoconvert ! glupload ! glcolorconvert ! filter_element_expecting_BGRA

Videoconvert gets to fixate these caps:

video/x-raw, format=BGRA; video/x-raw, format = {all, the, formats, glcolorconvert, supports}

When fixating however, videoconvert ignores the higher preference level expressed by having two separate structures in the caps, as it simply flattens all the pixel formats and picks one according to its scoring heuristic.

Ideally, videoconvert should consider each structure independently, unclear however whether this will break existing scenarios.