Should the optimizer avoid rewriting to `\W`?
See https://mathiasbynens.be/notes/es6-unicode-regex#impact-i.
Per ES6, /\W/iu was equivalent to /[^0-9a-jl-rt-zA-JL-RT-Z_]/u, and engines implemented it that way.
This was fixed in the spec in June 2016. Now, /\W/iu is equivalent to /[^0-9a-zA-Z_\u{017F}\u{212A}]/u. It has been fixed in the abovementioned engines, too, but some people are still using old versions.
TL;DR: inserting \W into character classes potentially changes the behavior of the regular expression, depending on the engine executing the optimized code.
Yeah, those K, S is a weird use case (bug). Might be an edge case, we'll see. Potentially yeah, will avoid rewriting \W (probably will be behind some option for legacy engines).