peekLF and peekSingleByteChar apparently aren't inlined, hurting the perf in this very hot loop. The perf regression is partly from function calls, and partly from the function depending on a closure. It also seems like there might be more gains to be had if I refactor this file to not depend so heavily on closures and nested anonymous functions.

Also count UTF-16be cr/lf as two bytes.

Just skip the file.

fix #16329