prod128: searchengine deployment issues

[sbesson]

See #448 (review)

• ansible/backup_searchengine_data.yml should be removed from this repository and added in the IDR submission workflow
• in its current implementation, ansible/restore_searchengine_data.yml will overwrite the search cluster with the state of the current backup if it exists

The second point might be mitigated by automatically creating a backup when running the indexer so that the backup and the elasticsearch cluster are not out of sync.
This suggests a more fundamental architecture issue related to the idempotency of these playbooks. If the searchengine index is already initialized, this initialization/restore task should be skipped altogether. Overwriting it should never happen in a regular workflow and should be a manual action.

Alternatively, as discussed during today's IDR meeting, it might be worth exploring whether ElasticSearch could use the same strategy as the PostgreSQL database i.e. cloning the volume containing the initialized cluster and having the application being able to start from this state. This would have the benefit of reducing the number of custom steps and points of failures to use a more standard workflow to copy the state of the indexer from one production deployment to another.