Discovered as the inefficiency behind https://github.com/r-lib/lintr/pull/2024 I think re_matches could be sped up substantially by doing better vectorization around non-match entries