regexp/syntax: canonicalization of letter classes breaks script selection

[DanielMorsing]

Program:

package main

import "regexp"

var r = regexp.MustCompile(`\p{Canadian_Aboriginal}`)

func main() {}

Output:

panic: regexp: Compile(`\p{Canadian_Aboriginal}`): error parsing regexp: invalid character class range: `\p{Canadian_Aboriginal}`

goroutine 1 [running]:
regexp.MustCompile({0x4b1f10, 0x17})
	/usr/local/go-faketime/src/regexp/regexp.go:313 +0xb4
main.init()
	/tmp/sandbox2354615464/prog.go:5 +0x1f

This was a program that compiled and ran on Go 1.24, but does not on any later version. The implementation of #70781 canonicalizes all strings within \p{} brackets before lookup. This causes selection of unicode scripts that contain underscores to no longer match.

[gopherbot]

Change https://go.dev/cl/747020 mentions this issue: regexp: apply canonicalization to scripts lookup

[mrkfrmn]

cc @rsc, @matloob