fix(regex): keep U+2028/U+2029 in Byte segments when Kanji mode is on

[maximilliangrand]

Bug

With Kanji mode enabled (a toSJISFunc supplied), a U+2028 (LINE SEPARATOR) or U+2029 (PARAGRAPH SEPARATOR) in the input is silently dropped, so the encoded QR decodes to different data than was passed in. No error is thrown.

const QRCode = require('qrcode')
const toSJIS = require('qrcode/helper/to-sjis')

const input = '皿a' + '
' + '晒三'  // Byte '皿a' + U+2028, then Kanji '晒三'
const { segments } = QRCode.create(input, { toSJISFunc: toSJIS })

// master: Byte segment bytes = [231,154,191,97]           -> U+2028 dropped, QR encodes '皿a晒三'
// fixed:  Byte segment bytes = [231,154,191,97,226,128,168] -> U+2028 preserved, full input encoded

Cause

The Byte-mode matcher in lib/core/regex.js is built as (?:.|[\r\n]). In JavaScript, . (without the s flag) matches every code unit except the four line terminators \n, \r, U+2028, U+2029. Only \r and \n were added back, so U+2028/U+2029 match no segment regex at all (numeric/alphanumeric/kanji don't cover them either) and are dropped during segmentation.

This exports.BYTE regex is only used for Byte segments when Kanji mode is on (see lib/core/segments.js). The default path uses the BYTE_KANJI negated class ([^A-Z0-9 $%*+\-./:]+), which does match those code points, so default mode is unaffected.

Fix

Add the two missing separators to the class:

-const byte = '(?:(?![A-Z0-9 $%*+\\-./:]|' + kanji + ')(?:.|[\r\n]))+'
+const byte = '(?:(?![A-Z0-9 $%*+\\-./:]|' + kanji + ')(?:.|[\r\n\\u2028\\u2029]))+'

Evidence

Added two test/unit/core/segments.test.js cases mirroring the existing \n case. They fail on master (not ok 25/not ok 26) and pass with the fix (26/26). Inspecting QRCode.create().segments directly confirms the Byte segment retains the U+2028/U+2029 bytes with the fix and drops them without it.