Add CNN-DM dataset to torchtext

[pmabbo13]

Description

Add CNNDM dataset to TorchText using TorchData datapipes

Process

1. Download CNN and DailyMail data from here. These downloads are cached, however the extractions from the tar files are not yet cached. See follow-up items.
2. Download URL lists from here, which designate which stories belong to the train/dev/test splits. These download are not yet cached. See follow-up items.
3. Use URL list to filter for stories that belong to target split.
4. Parse these stories to separate the article text from the abstract text and return a datapipe that yields (article, abstract) for each story.

Testing

Create mock dataset that mimics the format of raw CNNDM files and ensure that the datapipe yields the correct output on this mock dataset

pytest test/datasets/test_cnndm.py

Follow-Up Items

• Cache the extracted tar files for cnn_stories.tgz and dailymail_stories.tgz Cache CNNDM extraction and optimize reading in filenames #1809
• Cache URL lists (either on disk or using LRUCache) pending benchmark tests on how long it takes to instantiate them on the fly vs loading from cache Cache CNNDM extraction and optimize reading in filenames #1809

[Nayef211]

Overall the implementation looks great! Some followup items:

• Update the PR description

• Add a section in the description for followup items with a checklist of features that have yet to be implemented. We can then add a link to followup PRs that implement those features and check them off (similar to #1542)

• Add unittests for the dataset

• Let's try to remove new lines unless they add to readability of code. Here's what the guidance from pep8 style guide suggests

  Extra blank lines may be used (sparingly) to separate groups of related functions. Blank lines may be omitted between a bunch of related one-liners (e.g. a set of dummy implementations).

  Use blank lines in functions, sparingly, to indicate logical sections.

[Nayef211]

LGTM! Thanks for addressing all the PR comments. Let's wait for a review from @parmeet before we merge this in. 😄

[parmeet]

Thanks @pmabbo13 for adding this complex dataset to the repo. Overall it looks great. Just left few minor comments, but In general I think we should be good to land as such.