Change store() and update() to only store the canonical transcript dbID

if there is a canonical transcript attached to the gene.  If there
isn't, store a 0 (zero).  The canonical_transcript_id field does not
allow for NULLs.  This patch makes the unit tests happy.