Deduplicate resources for all datasets since the cost is neglible and it can only improve metadata quality.