Add Yandex, Google, and booru board search providers

[Patrick16]

Summary

Adds three new pluggable image search providers alongside DuckDuckGo, selectable per-job from a new Search source card in the UI, plus a SafeSearch control that now actually reaches all of them.

• Booru board (app/search/booru.py) — Danbooru-API-family boards: e621 / gelbooru / rule34 / danbooru. SafeSearch maps onto an explicit rating: tag instead of a hidden toggle.
• Yandex (app/search/yandex.py) — scrapes yandex.com/images. SafeSearch via a family cookie.
• Google (app/search/google.py) — scrapes Google Images, documented as unreliable (see below).

SearchConfig (app/models.py) + a build_search_provider() factory (app/search/__init__.py) replace the old hardcoded DuckDuckGoProvider() in jobs.py.

Live-verified findings (not just mocked)

• e621 works with zero configuration.
• Gelbooru / Rule34 now require an api_key + user_id — both returned HTTP 401 without one, a change from their historical open access. UI has fields for this.
• Danbooru is blocked by a Cloudflare bot-check even with a real browser UA, regardless of credentials, from at least some networks. Implemented (login + API key, the documented method) but best-effort. Also: anonymous Danbooru API access is capped at 2 combined tags.
• Yandex: confirmed the family cookie actually changes results (5/25 differed on a borderline query between family=0 and family=2).
• Google: confirmed it blocks plain HTTP scraping aggressively — instant HTTP 429 with full browser-like headers and no prior request history. Implemented as requested, but a block degrades to zero results + a log warning, not a crash, and the UI note says to expect this.

A real bug found and fixed along the way

While writing tests for the Yandex URL extraction, found that the original regex ([^&]+?) truncated any image URL that itself contained an HTML-escaped & (common in query strings) — it stopped at the first literal & instead of the real " terminator. Fixed to .+? (non-greedy up to the real terminator). Re-ran against the same live-captured Yandex page: recovered 5 more valid URLs (30 vs 25) that were previously silently dropped.

Tests

39 new tests across the new providers (tests/test_search_booru.py, tests/test_search_yandex.py, tests/test_search_google.py) and the factory (tests/test_search_factory.py); existing tests/test_jobs.py/tests/test_models.py updated for the build_search_provider() plumbing. Full suite: 166 passed.

Reviewer notes

• Google's provider is genuinely unreliable by nature of what it's scraping — that's expected, not a bug in this PR; it's surfaced clearly in the UI and README rather than hidden.
• Booru credential fields (booru_api_key, booru_user_id, booru_login) are persisted in the browser's localStorage like the existing LLM API keys, consistent with how this app already handles secrets (never written server-side).

🤖 Generated with Claude Code