fix: Optimize message reindexing to reduce sidekiq job creation (#12618)

Changes searchkick callback behavior to check `should_index?` before
creating reindex jobs, preventing unnecessary job creation for messages
that don't need indexing (activity messages, unpaid accounts, etc.).

Previously, `callbacks: :async` created reindex jobs for all messages
(~5,100/min or 7.3M/day in production), which were then filtered by
`should_index?` inside the job worker - resulting in 98% wasted jobs,
Redis memory pressure, and avoidable p0 alerts.

Now, `should_index?` is checked before job creation via `after_commit`
callback, reducing job creation to actual incoming/outgoing messages
from paid accounts.

  Changes:
  - Disable automatic searchkick callbacks
  - Add manual `after_commit` callback with `should_index?` condition
  - Add specs to verify callback behavior

  Expected impact:
  - 98% reduction in sidekiq job creation (~7.3M → ~150K jobs/day)
  - Reduced redis memory usage
  - Same async indexing behavior for eligible messages

app/models/message.rb

@@ -39,7 +39,7 @@
 #
 
 class Message < ApplicationRecord
-  searchkick callbacks: :async if ChatwootApp.advanced_search_allowed?
+  searchkick callbacks: false if ChatwootApp.advanced_search_allowed?
 
   include MessageFilterHelpers
   include Liquidable
@@ -135,6 +135,7 @@ class Message < ApplicationRecord
   after_create_commit :execute_after_create_commit_callbacks
 
   after_update_commit :dispatch_update_event
+  after_commit :reindex_for_search, if: :should_index?, on: [:create, :update]
 
   def channel_token
     @token ||= inbox.channel.try(:page_access_token)
@@ -436,6 +437,10 @@ class Message < ApplicationRecord
     conversation.update_columns(last_activity_at: created_at)
     # rubocop:enable Rails/SkipsModelValidations
   end
+
+  def reindex_for_search
+    reindex(mode: :async)
+  end
 end
 
 Message.prepend_mod_with('Message')