Evaluate approaches to handle deletions in the new reference

[tskir]

The algorithm in #16 has a caveat described in the Google Doc: if there's a large deletion in the new reference genome, the flanking sequences will fail to realign.

The doc also contains a sketch of a solution with using MSA for complete flanked allele sequences (both old & new) and extending the variant region to include the changes.