Add updated fuzzy search to benchmark

[a-johnston]

I benchmarked the v2 search with and without the short query optimization and as I was sorta expecting it didn't make a huge difference but it was slightly faster without sqo for the queries you had in the test:

Compared to the v1 Levenshtein distance approach, the top results were:

	v1	v2
sm	./entity/sword/sam.png	./junk/sam.png
ham	./entity/game_trap/ha_missed_me.wav	./entity/hamer/data.gd
push background	./entity/background_derelict/background/push.png	./entity/background_zone1/background/push.png

1 and 3 both used string length as tie breakers, 2 is debatable which should be the higher score but both are relevant. I also noticed that while on my older arch laptop the highlights were fine, on my windows and macbook the highlight was increasingly off presumably as scaling increases so I think the highlighting is missing accounting for something:

So I think the remaining things to do are

1. If the v2 results are good enough now, remove the v1 code
2. Rebase on the quick open changes
3. Rewrite the highlighting code and hopefully fix the above bug

[a-johnston]

Saved about a ms more by only allocating new result/match objects when needed. I tried rebasing against the upstream master and for some reason the file renames really mess with it so I just flattened everything into one commit with us as co-authors. If this branch's changes looks good to you we can move to that as the new flattened branch.