Align deletions after frame shift

[donkirkby]

#549 added references for SARS-CoV-2, but they have a strange frame shift part way through one of the gene regions. The code that aligns deletions to codon boundaries doesn't know about that frame shift, so it might report two partial deletions when the deletion is shifted to the wrong reading frame. See if deletions are very rare in SARS-CoV-2 and we can ignore this, or actually fix the deletion shifting.