feat: add VideoNet benchmark

[yadavta]

Summary

• Add VideoNet benchmark (5k videos, 4k test + 1k val multiple-choice questions, 4k binary questions)

Tasks added

Task	Description
videonet_mcq_test	test set for multiple-choice style of benchmark
videonet_mcq_val	validation set for multiple-choice style of benchmark
videonet_binary_0shot	binary style of benchmark with 0 in-context examples
videonet_binary_1shot	binary style of benchmark with 1 in-context examples
videonet_binary_2shot	binary style of benchmark with 2 in-context examples
videonet_binary_3shot	binary style of benchmark with 3 in-context examples

Details

• Data for CVPR 2026 Highlight paper titled "VideoNet: Domain-Specific Action Recognition in the VLM Era"
• Paper, HF Dataset, Website
• Please note that our benchmark has two different versions: a MCQ setup and a binary setup. I have separated the logic for these two setups, but am happy to refactor/combine if preferred.

Validation

• I've tested all tasks end-to-end on a few different servers that I have access to

Type of Change

• New benchmark/task

[yadavta]

Thanks for merging!