[Sskift]

UTF-16LE characters can contain the bytes of comment, quote or escape markers across adjacent code units. The parser currently recognizes those bytes as syntax even when no such character is present.

For example, \u2301\u0100 contains the bytes 23 00 at offset 1 and is truncated when comment: "#" is enabled. Similarly, \u2201\u0100 can cause an opening/closing quote error, and an unaligned custom escape marker can discard field bytes.

Match these markers only at UTF-16LE code-unit boundaries. The shared check uses the absolute buffered byte offset and current encoding, so it also handles encoding aliases, odd chunk boundaries and BOM discovery. UTF-8 and binary input take the existing unrestricted path directly.

This complements #545, which covers field and record delimiters and leaves other matchers outside its scope.

Validation:

• Comment regressions: eight failures and one passing binary control before the fix; all nine pass after it.
• Quote/escape regressions: twelve failures and one passing control before their fix; all thirteen pass after it.
• csv-parse suite: 582 passing, two pending, excluding the existing Windows sample-path collection failure tracked in fix: resolve sample file URLs as native paths #512.
• Additional independent checks cover encoding aliases, String/Buffer markers, BOM discovery, every two-chunk split, byte-sized writes and binary/other-encoding controls.
• TypeScript, changed-file ESLint/Prettier and git diff --check pass.

AI-assisted implementation and local verification using Codex.