Fix the ordering of transition callbacks:

Because pipeline status could be changed for the builds in the next
stages, if we process next stages first, the current build would be
out of synchronized, and would need a reload for that matter.

Alternatively, like what I did in this commit, we could process the
next stages later (by using `after_transition` rather than
`around_transition`), and complete what're doing for the current
build first. This way we don't have to reload because nothing is
out synchronized.

Note that since giving `false` in `after_transition` would halt the
callbacks chain, according to:

https://github.com/state-machines/state_machines-activemodel/blob/v0.4.0/lib/state_machines/integrations/active_model.rb#L426-L429

We'll need to make sure we're not returning false because we don't
intend to interrupt the chain.

This fixes #22010.

After this fix, both pipeline events and build events would only show
up once.

app/models/commit_status.rb

@@ -69,17 +69,15 @@ class CommitStatus < ActiveRecord::Base
       commit_status.update_attributes finished_at: Time.now
     end
 
-    # We use around_transition to process pipeline on next stages as soon as possible, before the `after_*` is executed
-    around_transition any => [:success, :failed, :canceled] do |commit_status, block|
-      block.call
-
-      commit_status.pipeline.try(:process!)
-    end
-
     after_transition do |commit_status, transition|
       commit_status.pipeline.try(:build_updated) unless transition.loopback?
     end
 
+    after_transition any => [:success, :failed, :canceled] do |commit_status|
+      commit_status.pipeline.try(:process!)
+      true
+    end
+
     after_transition [:created, :pending, :running] => :success do |commit_status|
       MergeRequests::MergeWhenBuildSucceedsService.new(commit_status.pipeline.project, nil).trigger(commit_status)
     end