[ice-pring]

Wrapping a long help description can split a valid UTF-16 surrogate pair across lines. For example, a formatter with width: 23 and a description of twenty repeated emoji produces a dangling high surrogate before the newline and a dangling low surrogate after it. Encoding that help output replaces the two halves with replacement characters.

Keep the long-word split boundary outside a surrogate pair. If an otherwise empty line has room for only one code unit, consume the whole pair so wrapping still advances. Remove an exhausted chunk and leave a full current line alone when the adjusted split has no room. Width accounting otherwise follows the existing code-unit approach.

Add eight port-specific cases covering odd boundaries, prefixed text, width one, a fractional width, unaffected ASCII/aligned cases, and the public format_help() output. Add a changelog entry.

Validation on Windows / Node.js 24.15.0 against 40086ea:

• Before the fix: 6 new cases fail and 2 controls pass. All eight pass afterward.
• npm test: ESLint and all 1,910 tests pass.
• An additional bounded check of 4,375 combinations of BMP/astral characters, spaces, and widths produces well-formed lines and preserves every non-whitespace character.
• git diff --check passes.

Other Node versions and operating systems were not run locally.