Oh interesting, I hadn't seen that. It looks like it uses the same idea of shuffle-lookups on the first three nibbles. That's a fairly large patch though, I don't think I've fully grokked it. At the very least, my code differs from that in that mine does less byte stream shifting, instead doing shifts in the scalar domain. Their code also needs to deal with quotes, escaping, etc. due to being part of a JSON validator.
Just today I stumbled on lemire works when looking through the dependencies of jgit and found the java ewah bitset implemenation and ended up browsing his code seeing the utf8 validator and simdjson. Coincedences..