add a simple script for video captioning

[sayakpaul]

Will pick up #34 but will repurpose it in the following manner:

• Take video files directly and leverage vllm to compute captions and serialize in a manner that is directly compatible with the new training API.
• The script will be simple and minimal, providing a sweet spot between performance and hackability.

I expect the script to be referenced by folks and repurposed instead of being a one-stop solution for all video models supported in vllm. We can iteratively refine it as we go. Planning to keep it under a directory called tools/.

This will also help GUIs a bit IMO: #116. Cc: @neph1.