I think the short answer is no, its not possible to distinugish poly-A(+) and poly-A(-) transcripts from normal total RNA-seq. Nor can you rely on things like lncRNAs to be non-polyA.
You could try a range of different lines of evidence to converge on a set of things you think are probably non-polyA.
You could start with the matched poly-A(+) and poly-A(-) data sets from ENCODE. If you do transcript specific quantification over something like GENOCDE or RNAcentral, and look for things with differential abundance between the + and - datasets.
You could then cross reference with polyA-seq data that specifically identifes polyA cleavage sites genome wide, and look for transcripst that don't have any signal, yet are highly expressed in the cell type in question.
Finally you could filter to only transcripts that didn't have a poly-A signal.