vout_subpictures: create subpicture regions from picture using their video_format_t

It has likely better information about the picture than a custom video_format_t on the side (mostly used for cropping and SAR signaling). In many cases it was the same format. In others (decoded image) it's better to use the format of the picture and modify a few values if needed.

This allows better separation of the region picture format and the region format (which may evolve in the future).

Also I fixed a regression from bbe2ee00