Add image-text-to-image and image-text-to-video tasks

[apolinario]

The goal of this new tasks is to support models that take in both image and text input and output either image or video.

The goal of this PR is making the tasks as analogous to image-to-image and image-to-video as possible, with the only difference that the image input should now be optional, as an empty image and a valid prompt should still work for a model like FLUX.2 (supports both text-to-image and image-to-image tasks) or LTX Video (both text-to-video and image-to-video)

Once this is in, I'll also have a widget PR in Moon to support this task in the model cards / widgets etc. and a follow up PR adding this to the inference providers, so that we can then PR repos to change the task for compatible models

[apolinario]

AI agent disclosure: I added the about.md files with Claude and haven't reviewed its slop yet. Will do it but should not be a blocker for the more structural stuff

[pcuenca]

cc @merveenoyan

[gary149]

looks good - maybe @merveenoyan you'll want to do some edits to md files

[merveenoyan]

thanks a lot for seeing this gap and working on it!

[merveenoyan]

would be nice to add more here and the Space too!

[Wauplin]

cc @julien-c as well on the (orthogonal) topic of generic any-to-any task + modality selection

[apolinario]

Thanks @merveenoyan , modified the examples and opened this PR here https://huggingface.co/datasets/huggingfacejs/tasks/discussions/12

[hanouticelina]

Reviewed the inference part, all good! thanks

[merveenoyan]

thank you!