Add enough functionality to parse the results of a BRAT experiment

[thcrock]

We would like to gather annotations using BRAT. We should add code to coordinate the BRAT installation.

This issue just tracks code to parse the results of an experiment. The code to start the experiment is being tracked in #136

This includes:

• Calculating inter-rater reliability
• Taking annotations from different users for the same job postings, merging them, and uploading the final skill candidates/labels to S3 somewhere

All of this will utilize S3. It is expected that a BRAT installation will somehow sync with s3, as opposing to making skills-ml code do a bunch of devops work to find and log into instances