fix(telegram): flatten Markdown horizontal rules in the sanitizer

Bare --- / *** / ___ HR lines confuse Telegram's legacy Markdown parser and (for ***/___) unbalance the delimiter count the sanitizer relies on, which causes the fallback to strip all formatting. Replace them with a plain Unicode divider (⎯⎯⎯) before the delimiter pass — same approach the bullet conversion already uses.

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

src/channels/telegram-markdown-sanitize.test.ts

@@ -65,4 +65,16 @@ describe('sanitizeTelegramLegacyMarkdown', () => {
   it('preserves indented list structure', () => {
     expect(sanitizeTelegramLegacyMarkdown('  - nested')).toBe('  • nested');
   });
+
+  it('flattens Markdown horizontal rules (---, ***, ___)', () => {
+    const input = 'before\n---\n***\n___\nafter';
+    expect(sanitizeTelegramLegacyMarkdown(input)).toBe(
+      'before\n⎯⎯⎯\n⎯⎯⎯\n⎯⎯⎯\nafter',
+    );
+  });
+
+  it('leaves horizontal rules inside code blocks alone', () => {
+    const input = '```\n---\n```';
+    expect(sanitizeTelegramLegacyMarkdown(input)).toBe(input);
+  });
 });

src/channels/telegram-markdown-sanitize.ts

@@ -28,6 +28,11 @@ export function sanitizeTelegramLegacyMarkdown(input: string): string {
   // as prose.
   text = text.replace(/^(\s*)[-+]\s+/gm, '$1• ');
 
+  // Flatten Markdown horizontal rules (bare --- / *** / ___ lines) to a
+  // plain Unicode divider. The parser doesn't understand HR syntax and the
+  // `*` / `_` characters would otherwise unbalance the delimiter counts below.
+  text = text.replace(/^[ \t]*[-_*]{3,}[ \t]*$/gm, '⎯⎯⎯');
+
   text = text.replace(/\*\*([^*\n]+?)\*\*/g, '*$1*');
   text = text.replace(/__([^_\n]+?)__/g, '_$1_');