First point: use an RNA-seq aware aligner such as Tophat, gsnap, STAR, mapsplice, etc.

Second point: All the variants reported by all variant-callers that I know of report the variants on the forward strand relative to the reference. The variant on the reverse strand is simply the reverse complement. If you want to define variants in RNA-seq in the orientation of the actual transcript, you will need to reverse-complement variants for all transcripts that are on the negative strand. So, a T-to-C site as reported by Varscan (regardless of the aligner) is going to be a T-to-C site on a forward-strand gene and the same site will be an A-to-G for a reverse-strand gene. In other words, you will need to post-process the variants as reported to get potential editing sites. As an aside, the strand filter in VarScan has nothing to do with whether the final called variant is on the forward or reverse strand; instead, it has to do with the presence of the variant in reads that align on only 1 strand or on both. Variants that are present in reads on only aligning to one strand are more likely to be false positives.

Third point: Unless you have DNA variants, these "editing" sites are just variants. There are plenty of A-to-G variants that are not editing sites, so you'll need to sequence the DNA in the same samples to determine whether or not editing might be happening. Though you don't mention it in your question, I assume you have an approach for doing that.