feat: anti-detection hardening, transaction cache, article parsing, structured write output

Anti-detection:
- Add 6 sec-ch-ua-* Client Hints headers (arch, bitness, full-version, etc.)
- POST requests now send Referer: x.com/compose/post + Priority: u=1, i
- follow/unfollow REST adds include_profile_interstitial_type param

Performance:
- Transaction ID cache with 1h TTL (~/.twitter-cli/transaction_cache.json)
- resolve_user_id: auto-detect screen_name vs numeric user_id

Features:
- Twitter Article parsing: extract long-form content as Markdown
- Write operations emit structured JSON/YAML when piped or OUTPUT env set
  ActionResult: {success, action, id, url, ...}

84 tests passing

twitter_cli/models.py

@@ -50,6 +50,8 @@ class Tweet:
     retweeted_by: Optional[str] = None
     quoted_tweet: Optional[Tweet] = None
     score: Optional[float] = None
+    article_title: Optional[str] = None
+    article_text: Optional[str] = None
 
 
 @dataclass