Improve ability to handle concurrent requests

[paulalbert1]

Top: current system for retrieval. Bottleneck is present when saving PubMed articles

Bottom: proposed design for making both retrieval and storage of PubMed articles using a bunch of "workers" (mostly likely EC2 instances).

Also, the "retrieval" call from ReCiter to PubMed service will be async. ReCiter will be notified of the success/failure of retrieval (list of pmids retrieved successfully, list of pmids retrieved unsucessfully) via a callback.

PubMed service will orchestrate the retrieval "workers" (retrieve PubMed articles) and storage "workers" (storing PubMed articles into DynamoDB). PubMed Retrieval Service still does what it does now: accept PubMed queries and returns results of the queries.

Q: How do you divvy up work among the workers? 100 each?
A: # pubmed_articles / # workers