New task: IR

[neubig]

This is not super-high priority, but it'd be nice to be able to analyze IR tasks. Some example benchmarks include

• MS MARCO: https://microsoft.github.io/msmarco/
• BEIR: https://github.com/UKPLab/beir

[pfliu-nlp]

Sounds good. A good idea is that we can start with our Dataset Finder dataset and models.