Restart the production redis fleet

Context: https://gitlab.com/gitlab-com/production/issues/431#note_96968283

For setting net.core.somaxconn to 1024 we need to do a rolling restart of the redis cluster. This should have no customer impact other than some sidekiq impact and a brief spike of 500 errors.

• Apply https://dev.gitlab.org/cookbooks/chef-repo/merge_requests/2445/
• Restart redis on the secondaries (redis-01 and redis-02)
• Wait for the synchronization process to finish in the redis log Synchronization with slave ... succeeded
• Restart redis on the primary (redis-03) to initiate the failover.