Sparse and robust normal and t- portfolios by penalized Lq-likelihood minimization