Use the translations of the _canonical_ transcripts of the source and target genes as the source of GO terms to project, rather than simply the longest translation in each, which was a bit arbitrary and didn't alway work.

Also make the print option for GO term projection a bit more verbose - aids debugging.