Towards robust databases: an ensemble-based workflow for error detection applied to chemical data