Remove hideInDatasets for multimodal tasks

[merveenoyan]

More and more datasets are showing up for multimodal tasks, and some authors are picking wrong task tags because hideInDataset is true, so removing them

[Vaibhavs10]

Do we have any intuition of how many such cases are there?

[pcuenca]

Yes, a few examples could be great for better understanding. I see, for example, this one that could possibly be assigned an image-text-to-text tag, but I wonder if other VQA datasets, such as the Cauldron, should have the same.

[pcuenca]

Spoke offline with Merve and had another look at things.

I'd be supportive of merging, given that:

• The tasks affected by this PR have non-empty datasets (image-text-to-text (90), any-to-any (13), visual-document-retrieval (8))
• video-text-to-text is already displayed in the filter bar
• The tasks here are incomplete (I'm not sure if they are populated from this file or not)
  
• @merveenoyan's intuition is that people pick from whatever options are offered/visible

But please, let's wait for vb to come back and see if he has additional insight!

[julien-c]

no objection!

[Vaibhavs10]

Thanks for pulling the numbers! Only recommendation/ suggestion would be to tag a few more datasets for the following:

any-to-any (13), visual-document-retrieval (8)

atleast so we have one page full of datasets.

[pcuenca]

Can we maybe merge this PR? We can always iterate later.

[merveenoyan]

sorry through the releases I couldn't work on this, @pcuenca I'm currently opening automatic PRs to a lot of models, I think it's ok to merge this

[merveenoyan]

I have opened more than 100 PRs, merging this, thanks a ton!