The short-circuiting of include_role in TaskExecutor was using the _task object
in the result dictionary. This causes the playbook class to be serialized, which
is something we no longer do due to speed/size problems with nested objects.
Simply removing this should fix a lot of memory and speed problems with include_role.