Observability and the inverse problem in electrocardiography