fix calculation bugs in encode_subframe_(): max_qlp_coeff_precision when qlp_coeff_pr...