FIX: properly log all internal job failures (#17805)

Our internal implementation of #perform on jobs performs remapping.

This happens cause we do "exception aggregation".

Scheduled jobs run on every site in the multisite cluster, and we report
one error per site that failed. During this aggregation we reshape the
context from the original object shape returned by mini_scheduler

The new integration test ensures this interface will remain stable even if
decoupled parts of the code change shapes.

Co-authored-by: Alan Guo Xiang Tan <gxtan1990@gmail.com>

lib/discourse.rb

@@ -177,6 +177,7 @@ module Discourse
 
     job = context[:job]
 
+    # mini_scheduler direct reporting
     if Hash === job
       job_class = job["class"]
       if job_class
@@ -184,6 +185,11 @@ module Discourse
       end
     end
 
+    # internal reporting
+    if job.class == Class && ::Jobs::Base > job
+      job_exception_stats[job] += 1
+    end
+
     cm = RailsMultisite::ConnectionManagement
     parent_logger.handle_exception(ex, {
       current_db: cm.current_db,