Fix nasty bug that would come up only if a client connection to a remote

ldap server suddenly dies.
We were creating a wrong talloc hierarchy, so the event.fde was not
freed automatically as expected. This in turn made the event system call
the ldap io handlers with a null packet structure, causing a segfault.
Fix also the ordering in ldap_connection_dead()
Thanks to Metze for the huge help in tracking down this one.

source4/ldap_server/ldap_bind.c

@@ -208,6 +208,7 @@ static NTSTATUS ldapsrv_BindSASL(struct ldapsrv_call *call)
 		} else {
 			ctx->conn = conn;
 			status = gensec_socket_init(conn->gensec, 
+						    conn->connection,
 						    conn->connection->socket,
 						    conn->connection->event.ctx, 
 						    stream_io_handler_callback,