Use Elasticsearch for autocomplete_sources/members

Implement backend changes required to make Issue and MR notes use Elasticsearch search-as-you-type.

Context

Issue and MR note @mentions are notoriously slow when searching for a user for /assign, /assign_reviewer, /cc, etc. An initial load makes an API call to fetch all possible users and store them in cache for subsequent searches. For the gitlab project, a few thousand users are returned and cached making the initial call take 3.37 seconds to receive a response from the backend.

We want to move to using Elasticsearch to provide autocomplete results for users. The first feature we want to change is @mentions on Issue and MR notes but this could be extended to other features as well.

#388197 is the frontend part of this backend issue.

Implementation

Make use of the User index and Elasticsearch search-as-you-type to provide a more responsive autosuggest experience on Issue and MR notes.

Requirements:

• Changes are behind a feature flag autosuggest-users-using-elasticsearch and guarded by Elastic::DataMigrationService.migration_has_finished? for create_user_index and backfill_users.
• Implement a way to make an elasticsearch call to fetch matching users when use_elasticsearch? is true.
• Update the mappings of username and name to allow for search-as-you-type.
• Return users that match the term AND are one of:
  • The note owner
  • A participant of the note
  • A member of the project: get this as part of the Elasticsearch query using namespace_ancestry_ids since the database query for this takes 0.5 seconds.
  • @all or one of current_user.authorized_groups: for the first iteration this could be done as a separate step using database calls and future iterations could see these options added to elasticsearch.
• If possible, use the busy field from elasticsearch to remove the need for lazy loading availability
• Sorting: The owner and participants of the note are shown at the top, then users, then @all and groups