Optimised paused runners to not re-query every 3s

We have long polling mechanism which is not being used by paused runners.
This makes the paused runner to generate a request every 3s,
where it could be fired only every 1h.

changelogs/unreleased/optimise-paused-runners.yml

+---
+title: Optimise paused runners to reduce amount of used requests
+merge_request:
+author:
+type: performance

lib/api/runner.rb

@@ -84,7 +84,11 @@ module API
       end
       post '/request' do
         authenticate_runner!
-        no_content! unless current_runner.active?
+
+        unless current_runner.active?
+          header 'X-GitLab-Last-Update', current_runner.ensure_runner_queue_value
+          break no_content!
+        end
 
         if current_runner.runner_queue_value_latest?(params[:last_update])
           header 'X-GitLab-Last-Update', params[:last_update]

spec/requests/api/runner_spec.rb

@@ -351,11 +351,13 @@ describe API::Runner, :clean_gitlab_redis_shared_state do
       context 'when valid token is provided' do
         context 'when Runner is not active' do
           let(:runner) { create(:ci_runner, :inactive) }
+          let(:update_value) { runner.ensure_runner_queue_value }
 
           it 'returns 204 error' do
             request_job
 
-            expect(response).to have_gitlab_http_status 204
+            expect(response).to have_gitlab_http_status(204)
+            expect(response.header['X-GitLab-Last-Update']).to eq(update_value)
           end
         end