[boyan01]

Summary

Stacked on #544. Double clicking Chinese text selected one character because Flutter's bundled ICU has no CJK dictionary. Word selection now asks word_segmenter (#548) first.

 markdownWordRangeAt(text, offset, upstream)
   run = characters between whitespace / separators (incl. CJK punctuation)
   if run is ASCII → return run
-  TextPainter.getWordBoundary(run, index)        # 我们 → 我
+  wordBoundaries(run)?.segment containing index   # CFStringTokenizer / system ICU / Intl.Segmenter
+  ?? TextPainter.getWordBoundary(run, index)      # no platform segmenter

• New dependency: word_segmenter: ^0.1.0 (already on pub.dev; resolves to the workspace package here).
• ASCII runs and separator handling are unchanged.

Evidence

Before: the strengthened markdown_word_selection_test.dart against the old lib/:

Chinese text breaks at dictionary words           Expected: '我们'  Actual: '我'
double click selects a word in a Chinese paragraph Expected: '效果'  Actual: '效'

After: 5/5. 我们使用Flutter开发桌面应用 selects 我们 / 使用 / Flutter / 开发 / 桌面 / 应用 at every offset and from the right half of each word's last character; double click on 效 selects 效果, on 桌 selects 桌面.

Platform	flutter test (package)
macOS 27	267 passed
Windows (remote, system icu.dll)	267 passed
Ubuntu 22.04 (remote WSL, system libicuuc)	267 passed

flutter analyze: no issues. Not checked by hand in the GUI yet.

Merge Danger

Door: two-way

Blast Radius: selection

Only double-click word selection in non-ASCII runs changes. Segments come from platform dictionaries, so the exact word can differ by platform (e.g. macOS splits 很好 into 很 / 好). Platforms without a segmenter (Android < 31, very old Windows) keep the previous per-character behavior. Releasing the next mixin_markdown_widget adds word_segmenter to consumers' dependency trees.