We support round-tripping:
- 24-bit RGB
- 32-bit RGBA (new)
- 8-bit grayscale (new)
- yuv420p (new)
- yuyv422 (new)
We drop the gray16 conversion code as it was not handling byte order correctly, and could not be triggered anyway due to an impossible test:
frame.format.name == ('gray16le', 'gray16be')