Just look for polyA tracts at the end of sequences, and count them if they're larger than ~18 bases. I've done this with MinION cDNA reads by mapping the polyA adapter sequence (with an elongated polyA sequence) to the reads using LAST, and working out the length of aligned sequence.
Unfortunately, because the cDNA adapter finishes with a TVN sequence, rather than TTT, binding of the adapter is not necessarily a good representation of the true polyA length. I expect that Illumina adapters will have similar problems, if not worse (due to the read length limits).
It might be possible to use direct-RNA sequencing to better categorise the polyA lengths, but I've yet to look at the public direct-RNA datasets. Just in case you're interested in looking yourself, here's a direct RNA dataset for NA12878 from the wonderful nanopore-WGS consortium:
Here's the consortium's graph from that summary document showing polyA length: