Interesting. These sizes appear correct (below) and using 3.2.1, we know this video is 720p. 1280x720 is what V4L reports as well. You mentioned the metadata lines above, but I assumed that was a cut and paste error from the original post. Should I try 4 lines as shown there?
root@nvidia-desktop:~# v4l2-ctl --list-formats-ext
ioctl: VIDIOC_ENUM_FMT
Index : 0
Type : Video Capture
Pixel Format: 'UYVY'
Name : UYVY 4:2:2
Size: Discrete 1920x1080
Size: Discrete 1280x720
Size: Discrete 720x289
Size: Discrete 720x245
Size: Discrete 1920x541
Size: Discrete 640x512
Size: Discrete 640x480
Size: Discrete 1280x1024
Index : 1
Type : Video Capture
Pixel Format: 'NV16'
Name : Y/CbCr 4:2:2
Size: Discrete 1920x1080
Size: Discrete 1280x720
Size: Discrete 720x289
Size: Discrete 720x245
Size: Discrete 1920x541
Size: Discrete 640x512
Size: Discrete 640x480
Size: Discrete 1280x1024
Index : 2
Type : Video Capture
Pixel Format: 'UYVY'
Name : UYVY 4:2:2
Size: Discrete 1920x1080
Size: Discrete 1280x720
Size: Discrete 720x289
Size: Discrete 720x245
Size: Discrete 1920x541
Size: Discrete 640x512
Size: Discrete 640x480
Size: Discrete 1280x1024
I know some of these sizes are odd looking, but they are correct, for our use case. SDI has interlaced options, which we transport over the CSI as two “frames (fields)” and recombine the fields to get a full frame.