For now this is just a standard implementation of the longest
common subsequence algorithm over the lines, except that it doesn't
do any coalescing of the lines. This isn't really ideal since
we get a single Hunk per changed line, and is definitely something
to improve in the future.