Fix UTF-16 column counts in TextLength - #345715

Fixes #345714.

TextLength.ofText undercounts supplementary Unicode characters because its string iterator yields code points while editor columns use UTF-16 code units. Inserting an emoji and applying the inverse edit leaves a lone surrogate; trimming a completion's common prefix can delete an existing character at a secondary cursor. Count each yielded string's UTF-16 length so inverse ranges and multicursor completions preserve the original text.

Added regressions for UTF-16 length calculation, multiline and multiple-edit inverses, common prefix/suffix reduction, and the actual getSecondaryEdits path. All seven new regressions fail before the fix and pass afterward.

Validation on macOS arm64, Node.js 24.22.1, pnpm 12.16.0:

• pnpm run build-fast --client-only: passed.
• pnpm run test-node --runGlob 'vs/editor/test/common/core/{textEdit,textLength}.test.js': 127 passed, 2 existing skipped fuzz tests.
• pnpm run test-browser-no-install --browser chromium-chrome --run src/vs/editor/contrib/inlineCompletions/test/browser/getSecondaryEdits.test.ts: 4 passed using local Chrome.
• pnpm run eslint on all four changed files: passed.