A BSDE-based approach for the optimal reinsurance problem under partial information