Add web_split dataset for Paraphase and Rephrase benchmark

## Describe:
For getting simple sentences from complex sentence there are dataset and task like wiki_split that is available in hugging face datasets. This web_split is a very similar dataset. There some research paper which states that by combining these two datasets we if we train the model it will yield better results on both tests data.

This dataset is made from web NLG data.

All the dataset related details are provided in the below repository

Github link: https://github.com/shashiongithub/Split-and-Rephrase