Rudimentary statistics for the dataset

[zouharvi]

This PR adds a script to compute some basic statistics about the corpus and displays them in the README. It is as much as I can extract from the fields that are currently available. From NLP perspective I'd be mostly interested in the languages and number of sentences. Let me know if you have some more ideas about this.

Semi-related notes:

• I also updated the total row count based on what's in data/acl-publication-info.74k.parquet.
• A more standard approach to TODOs would be to use GitHub Issues and list them there, instead of having it in the README. This would make tracing progress easier and would also make the "title page" less cluttered. What do you think?
• S2 provides paper embeddings. We could do some fun stuff with that.

[shauryr]

1. about the todo - I agree. I will remove the todo section soon and more it to issues
2. yeah I have been working with - https://huggingface.co/sentence-transformers/allenai-specter : trying to map things and find clusters