increase the MASTER_CONNECT_RETRY time under valgrind,
otherwise the slave gives up retrying before the master is ready

also, cosmetic cleanup of rpl_semi_sync_master_shutdown.test