We're using 3.0.14 in prod, test with that

Install newrelic_rpm to < 4.0

NewRelic have dropped support for ruby 1.9.x, and so we can't use their latest
app monitoring agent until we upgrade ruby.

ref https://discuss.newrelic.com/t/support-for-ruby-jruby-1-x-is-being-deprecated-in-ruby-agent-4-0-0/44787

TNL-6562: Enhance rake logging.

Verify that ES is configured properly before starting

Run the rake search:validate task before starting the service. If it
fails, then that means the mappings or indexes aren't in sync, the
former of which can cause corrupt data, and the latter, non-functioning
search. It fails with this exit code so that supervisor knows whether to
sleep before trying to restar the app or not.

TNL-6447: Make search:initialize idempotent.

- Make search:initialize idempotent.
- Add rake tasks for put_mappings and validate_index.

TNL-6464: Minor changes to rebuild_index.

- Default to moving alias and running catchup.
- Allow configurable extra minutes for catchup.

Better filtering for rake search:initialize behavior

Diana/elasticsearch update

- Using sinatra-param to validate parameters instead of waiting for search failure
- Consolidated some code with shared_examples
- Add more logging around Search

TNL-6118: Replaced Tire with elasticsearch-model (newest)

- Replaced Tire with elasticsearch-model
- Updated README
- Added task to initialize a new index
- Fix and add tests
- Remove search:reindex and search:prune in favor of search:reindex
- Use milliseconds in auto-generated index name.
- Delete unused application.rb.

TNL-6118: Fix rebuild by loading app.rb in Rakefile

Fix typo in search:do_reindex task

use 16.04 latest docker image for forums

The core issue was that if you don't provide an install step travis
tries to do its default action which generates a bunch of ruby code
that is linked to the ruby on the host system instead of the ruby in
the container.

Add a default response size and a response size limit.

By default, we currently allow a user to retrieve all responses from a
thread, which is terrible from a performance standpoint.

This change adds a default response size of 100, which was chosen to
match typical production pagination parameters, and a response limit of
200.  Given that responses can include children comments, 200 is still a
large number, but far less disastrous than, say, getting 3000 responses,
with all of their children comments.

Both of these values are configurable, and can be overriden via
environment variables.

Add method tracing inside user.

TNL-5115: Remove sort order from cs_comment_service

MA-2139: Mark thread as read for leaner on thread/comment creation

Add an OEP-2 compliant openedx.yaml file