De-identification of Free Text in Swiss-French Electronic Health Records: From Internal Annotation Crowdsourcing to Personal Data Detection