users: Fix ordering issue with deactivating bots.

The new comment explains the issue in some detail, but basically if we
deactivate the bots first, then an error partway through is corrected
by a retry; if we deactivate the user first, then we may leak
undeactivated bots if a failure occurs.

zerver/lib/actions.py

@@ -1146,6 +1146,16 @@ def do_deactivate_user(
     if not user_profile.is_active:
         return
 
+    if _cascade:
+        # We need to deactivate bots before the target user, to ensure
+        # that a failure partway through this function cannot result
+        # in only the user being deactivated.
+        bot_profiles = UserProfile.objects.filter(
+            is_bot=True, is_active=True, bot_owner=user_profile
+        )
+        for profile in bot_profiles:
+            do_deactivate_user(profile, _cascade=False, acting_user=acting_user)
+
     with transaction.atomic():
         if user_profile.realm.is_zephyr_mirror_realm:  # nocoverage
             # For zephyr mirror users, we need to make them a mirror dummy
@@ -1200,13 +1210,6 @@ def do_deactivate_user(
         )
         send_event(user_profile.realm, event, bot_owner_user_ids(user_profile))
 
-    if _cascade:
-        bot_profiles = UserProfile.objects.filter(
-            is_bot=True, is_active=True, bot_owner=user_profile
-        )
-        for profile in bot_profiles:
-            do_deactivate_user(profile, _cascade=False, acting_user=acting_user)
-
 
 def do_deactivate_stream(
     stream: Stream, log: bool = True, acting_user: Optional[UserProfile] = None