[WIP] Data rearchitecure proof of concept

[gabina]

What this PR does

This PR makes a small proof of concept for the data re-architecture project. Basically:

• It adds a new RevisionDataManager class that given a course, a wiki and a specific period of time (in the future, a timeslice), it fetches revision and revision scores from different APIs, and returns an array with all that data.
• It adds the client of the RevisionDataManager class: UpdateCourseStatsTimeslice class that updates timeslices from this revision data in RAM (not commited yet).
• It implements the update_cache_from_revisions method for the ArticleCourseTimeslice, which updates an ArticleCourseTimeslice record with the new revisions in memory.
• It updates the existing update_cache method for the ArticleCourse, to update caches based on the ArticleCourseTimeslice records instead of the revision table (ArticlesCourses.update_cache_from_timeslices).

This PR adds specs for all the new behaviors.

Open questions and concerns

There are a lot of TODO comments that will be handled in the future. A similar process should be done for CourseWikiTimeslice and CourseUserWikiTimeslice timeslices.

For ArticlesCourses records, now view_count is calculated based on the first non-empty article course timeslice record start date (instead of using the first revision). We estimate the first non-empty record checking user_ids field is not null. This should lead to the same results.

[gabina]

RevisionScoreImporter specs are failing, I think I understand why. Will fix it on Monday.

[gabina]

The following specs are failing but they're not related to my changes. I think both specs were already fixed in master. I'll sync my own data-rearchitecture-for-dashboard branch with master in a next PR.

rspec ./spec/lib/replica_spec.rb:151 # Replica API requests functions identically on non-English wikis
rspec ./spec/lib/replica_spec.rb:98 # Replica API requests Article retrieval when fetching existing articles returns article "Autism"