Improve algorithm requirements scoring #18
Description
Right now the algorithm is using a hard-to-compare/combine [0, 1] scoring for requirements. It suffers from an apples-to-oranges comparison when comparing different requirements, since each one has different distribution in that range, which means some are unintentionally weighted heavier than others.
It will probably be better, and approach a correct solution faster, if we sort primarily on a "minimum substitutions to meet requirement" heuristic. This gives a more concrete value, and allows more even comparisons between scores. To aggregate across requirements, we could average the minimum substitutions - summing seems too large, since 5 requirements with a minimum substitutions of 1 could all be resolved in a single substitution.
The existing weighting could be used as a secondary sort criteria for matching min substitutions scores. We'll probably want to figure out how to only compare like to like (only compare the same requirement types), but I'm exactly sure what that will look like yet.