fix: avoid splitting surrogate pairs when truncating wide characters

[greymoth-jp]

truncateWidth takes a fast path when str.length === strlen(str):

if (str.length === strlen(str)) {
  return str.substr(0, desiredLength);
}

The idea is "every character is one code unit and one column, so cutting by code unit is safe." That holds for ASCII, but it's also true for surrogate-pair characters such as CJK Extension B (e.g. 𠮷, used in the surname 𠮷田) or emoji: they are two code units and two columns, so length and strlen stay equal. substr (and the slice(0, -1) loop underneath) cut by code unit, so truncating on the boundary of such a character leaves a lone surrogate:

const { truncate } = require('cli-table3/src/utils');
truncate('a𠮷bc', 3); // => "a\uD842…"  — lone high surrogate, renders as the replacement char

In a table this shows up as a � in the cell whenever a wide character lands on the truncation point.

The change keeps the fast path for plain strings, skips it when a surrogate is present, and trims by code point in the slow path so a wide character is never split. BMP input (including the existing full-width CJK cases) is unaffected. Added tests for a CJK Extension B character and an emoji.