Avoid splitting surrogate pairs when shortening address labels

[greymoth-jp]

The address-label shortening loop in modules/svg/labels.js trims one character at a time using substring with a UTF-16 .length count:

name = `${name.substring(0, name.replace(/…$/, '').length - 1)}…`;

.length counts UTF-16 code units, so length - 1 can land in the middle of a surrogate pair. When the label ends in a character outside the BMP, that character is two code units, and substring keeps only the first half. The result is a lone surrogate that renders as �.

This is reachable for real address points. utilDisplayName resolves the label from name, then addr:housename, then the house number, and both name and addr:housename are free text. A house or building name ending in an emoji, or in a CJK Extension B ideograph such as 𠮷 (U+20BB7, which appears in Japanese names), hits this case once the label is wide enough to be trimmed.

The fix uses Array.from(...).slice(0, -1) to drop the last code point instead, the same approach utilUnicodeCharsTruncated already takes in modules/util/util.ts. Strings within the BMP, including all BMP CJK, are unchanged.

Checked with node:

"ビル𠮷"      old: "ビル\uD842…"  (lone surrogate)   new: "ビル…"
"漢字テスト"   old: "漢字テス…"                        new: "漢字テス…"  (unchanged)

I did not add a test, since this runs inside the canvas label-rendering loop. utilUnicodeCharsTruncated, which uses the same technique, is covered in test/spec/util/util.js.

[matkoniecz]

the same approach utilUnicodeCharsTruncated already takes in modules/util/util.ts

can utilUnicodeCharsTruncated be used here directly?