The accuracy of the general practitioner's sense of alarm when confronted with dyspnoea and/or thoracic pain: protocol for a prospective observational study