Cache column_exists? for Elasticsearch columns

Any time a searchable object (e.g. issue, merge request, project, etc.) is updated, the after_commit hook checks whetether two Elasticsearch columns are present in the application_settings table. This leads to unnecessary queries in production as the value can just be cached by connection.

Closes gitlab-org/gitlab-ce#43355

CE port: https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/17780