add multi gpu image captioning

[dame-cell]

What does this PR do?

This PR adds a distributed image captioning example using blip.

Based on https://gist.github.com/sayakpaul/dfb26d6562b62ba8c2f613dc912a28a8#file-generate_captions_multigpu-py

This was discussed on the #3078

Who can review?

@sayakpaul

Notes

• I was able to generate image captions using two T4 GPUs in less than 20 minutes for a dataset of 50,000 rows.
• If there are any concerns regarding the choice of model or dataset, I am open to making adjustments as needed.

[sayakpaul]

Thanks for your contributions! But we already have #3123. Does it not have an overlap in the sense that it also adds captioning?

[dame-cell]

oh my bad i thought that was like for inference with florence

[dame-cell]

@sayakpaul i guess the only thing left if for the Speech data generation i will try contributing this if no one has done it

[sayakpaul]

Sure, speech-data generation would be very nice.