When searching for merge requests in the context of a group we don't
need to use an Elasticsearch join. The join logic [defined in
ApplicationClassProxy](
https://gitlab.com/gitlab-org/gitlab/-/blob/339e55b484ce7f4250ef5bbb451c9ed2eac7c6e2/ee/lib/elastic/latest/application_class_proxy.rb#L145
) is not necessary as we are already filtering down based on the project
ID list the user has access to and hence we can just use the
`target_project_id` field instead of joining to the project and using
the ID field.

This MR is basically identical to
https://gitlab.com/gitlab-org/gitlab/-/merge_requests/46461 but in this
case we need to use `target_project_id` instead of `project_id` since
the MR only had `target_project_id` until recently and `project_id` is
not yet backfilled.

This change on it's own will make searches more performance (since this
term query is more efficient than a join) but it will additionally be
the next step in allowing us to move merge requests to a separate index
in Elasticsearch https://gitlab.com/groups/gitlab-org/-/epics/5468