feature proposal: heaviest common subsequence algorithm for linear token diffs

[dgetu]

I saw that this wiki page describes a problem with the LCS approach to linear token diffs. This paper seems to have a solution for the related HCS problem. If tokens are weighted by the number of characters they contain (maybe with keywords at a reduced weight and whitespace at zero weight), I'd think that would result in good diffs in most cases. Per-language weight tuning would be possible as well.