feat(masters): add --add-sitelink/--remove-sitelink to masters update

[axisrow]

Summary

Implements быстрые ссылки (sitelinks) support in direct masters update — part of issue #648's Этап C (structural sections: sitelinks, audience, Metrika counters/goals). This is a browser-driven feature (Мастер кампаний has no API), following the same add/remove-by-position pattern already established for --add-audience-tag/--remove-audience-tag (#681), adapted for sitelinks' 3-field card structure (Title/Href/Description) instead of a plain text tag.

• --add-sitelink "Title|Href|Description" — pipe-delimited, adds a new card via the "Добавить" button.
• --remove-sitelink POSITION — removes a card by its 0-based on-page position.
• _read_sitelinks/_add_sitelink/_remove_sitelink in direct_cli/browser/masters.py, wired into update_master (snapshot-before + high-to-low removal, mirroring the audience-tags convention) and _verify_saved (multiset check on (Title, Href) — Description is not bulk-readable, see below).

⚠️ NOT LIVE-VERIFIED

Testid structure was confirmed via two live read-only recon passes against campaign 713277109 (Playwright + mcp claude-in-chrome, no mutations saved) — but the actual add/remove/save behavior has not been exercised end to end against a real Yandex session. Specifically open questions:

1. How the inline SitelinkRow.* edit form closes/commits — no dedicated "Готово"/save testid appeared in recon. The code presses Escape then clicks the section container as a best-effort dismiss, relying on update_master's own re-read (_verify_saved) as the real proof of success rather than trusting the close action itself.
2. contenteditable vs plain <textarea> for SitelinkRow.name/href/description.textarea — conservatively assumed contenteditable (same class as headline/text slots), reusing _clear_text_field + .type(). If it turns out to be a plain textarea, this degrades to a harmless no-op equivalent, not a functional bug.
3. Whether "Добавить" opens a new card with the SitelinkRow.* triple immediately available — assumed by analogy with every other "click Добавить, fill, commit" flow in this module (target actions, audience tags), never itself clicked in recon.
4. The real maximum sitelink count — _SITELINKS_SLOT_COUNT = 5 comes from UI copy ("добавить до 5 штук"), not from actually adding a 5th sitelink past the 4 the recon campaign already had.

None of this is claimed as working in the docstrings/CLI help — see the module comment above _SITELINKS_EDITOR_TESTID in direct_cli/browser/masters.py for the full recon writeup. A live mutation-verification pass is planned as a follow-up (browser session currently queued for other in-flight work) before this is considered safe to rely on in production.

Test plan

• pytest tests/test_masters.py -k "sitelink or Sitelink" — 28 passed
• Full offline suite pytest -n auto — 3405 passed, 23 skipped, no regressions
• black/flake8 clean on all changed files
• Live verification (add, remove, save-persists) — pending, tracked as a follow-up before this is relied upon

Relates to #648 (no dedicated sub-issue for this specific section).

🤖 Generated with Claude Code

https://claude.ai/code/session_01FF5MKQjSX89UpwHfFG3uBe

[axisrow]

Закрываю и переоткрою для форсирования нового CI-прогона — GitHub Actions застрял в queued больше 1.5 часа с рассинхроном Checks/Actions API, новые пуши (пустые коммиты) не триггерят новый pull_request synchronize event.

[axisrow]

Закрываю окончательно и пересоздаю как новый PR — CI застрял без единого нового workflow run уже несколько часов, включая после push и close/reopen. Новый PR должен получить чистый webhook-цикл.