fix(terminal): stop link hover regex from swallowing trailing CJK text

The clickable-link regex in TierTerminal used [^\s<>()"'] which lets the
match keep eating non-ASCII chars after a URL, so a sentence like
"https://x.json，看到一大堆 JSON" highlighted the URL plus all the
trailing Chinese as one giant link.

URLs are ASCII per RFC 3986. Adding the U+0080–U+FFFF range to the
negated set makes the regex stop at the first non-ASCII char without
breaking Windows file paths (those keep the looser set so D:\桌面\foo.txt
still matches).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: claude

src-ui/src/components/center/TierTerminal.tsx

@@ -522,7 +522,11 @@ function TierTerminalImpl({
     // Underlines matched tokens on hover; click opens via Tauri's open_url
     // command (delegates to the OS shell — system browser for URLs, default
     // handler for local files like report.html).
-    const LINK_RE = /(https?:\/\/[^\s<>()"']+|file:\/\/\/[^\s<>()"']+|[A-Za-z]:[/\\][^\s<>()"']+)/g;
+    // URLs are ASCII per RFC 3986; the €-￿ guard stops the match at
+    // any non-ASCII char so trailing CJK punctuation/text (e.g. "https://x，看到…")
+    // doesn't get swallowed into the link. File paths keep the looser set so
+    // Windows paths with Chinese folder names still match.
+    const LINK_RE = /(https?:\/\/[^\s<>()"'€-￿]+|file:\/\/\/[^\s<>()"'€-￿]+|[A-Za-z]:[/\\][^\s<>()"']+)/g;
     term.registerLinkProvider({
       provideLinks(bufferLineNumber, callback) {
         const line = term.buffer.active.getLine(bufferLineNumber - 1);