feat: add SerpBase (Google) search engine via REST API

[gefsikatsinelou]

Summary

Adds a third SearchEngine backend, method="serpbase", which fetches Google results from the SerpBase REST API instead of driving a browser or scraping HTML. Results are rendered into a minimal Google-style SERP document so they flow through the existing parser unchanged — you get the usual general results with the standard schema.

Background

Plan 051 (v0.11.0) kept requests as the no-browser option but noted it is "increasingly blocked for live collection". The browser path (patchright) works but needs an installed Chrome and a display (Xvfb on headless hosts), and both paths hit Google's /sorry/ CAPTCHA wall on shared/VPS IPs. This PR adds a pure-HTTP backend that doesn't scrape at all: it calls the SerpBase Google Search Results API (GET https://api.serpbase.dev/google/search?q=...&api_key=...&num=...) and renders the JSON into the classic #rso > div.g > div.yuRUbf structure the parser already handles.

Changes

• New: WebSearcher/searchers/serpbase_searcher.py — SerpBaseSearcher with the same send_request/cleanup contract as the other backends, plus a _json_to_html renderer mapping organic_results (title/link/snippet) to minimal Google-style markup.
• Updated: WebSearcher/models/configs.py — SearchMethod.SERPBASE and SerpBaseConfig (api_key, base_url, timeout), wired into SearchConfig.
• Updated: WebSearcher/searchers/searchers.py — SearchEngine.__init__ accepts serpbase_config and dispatches method="serpbase".
• Updated: WebSearcher/demos/cli.py — ws-demo search/searches accept serpbase.
• Updated: README.md — short "API method (no browser, no scraping)" section.
• Updated: tests/test_searchers.py — no-key graceful path + end-to-end JSON → parsed-results test.

Design decisions

Decision	Rationale
API key via serpbase_config / SERPBASE_API_KEY env var	Matches the existing per-backend config pattern (requests_config, patchright_config)
Graceful skip when key missing	Empty ResponseOutput (response_code 0) + warning — a crawl without a key doesn't crash or block the other methods
Render JSON to minimal HTML instead of a parallel result schema	The parser/classifier pipeline stays the single source of truth; saved records are identical in shape to browser/requests methods
Reuses requests	No new dependencies

Testing

• No key set: logs a warning, returns an empty response, no exception.
• Key set (via env or config): parse_serp produces general results with the correct title/url/text.
• Existing tests pass unchanged (12 passed).

Related

• Plan 051's note that the requests path is "increasingly blocked for live collection" — this is a pure-HTTP path that doesn't scrape.