vout_subpictures: separate the region picture from its placement

Each region contained a video_format_t and a picture with its own video_format_t, which any clear indication of which is supposed to be used for what.

Now the picture has its own video_format_t which is used as any other picture. And there's a subpicture_region_crop structure replacing the other video_format_t that only contains the bits that are used to crop the region within its parent video. The SAR of the crop has no use in display modules.

There's a drawback compared to the old video_format_t, it doesn't contain colorimetry information for text regions. But since noone was setting these values, it makes no difference. We may add them later if we want to render HDR text.

Tested with text, OSD, DVD menus on D3D11, D3D9, OpenGL and WinGDI.

On top of !4898 (merged), !4874 (merged) and !4888 (merged)