Add SerpBase Google search modes - #46

🔍 Problem

• SerpBase users cannot use their existing API key through webfox, including Google's specialized search endpoints.

🛠️ Solution

• Cover all six endpoints through the existing search capability and one web_search tool: search, images, news, videos, maps, and maps-detail.
• Expose typed native options and mode-aware validation/defaults. Maps Detail takes feature IDs from Maps Search in queries, without dummy queries or additional tools.
• Include media URLs, sources, feature IDs, contact information, and place details in model-visible text; retain richer metadata in JSON. Connect provider guidance to tool descriptions.
• Keep each input to one page or place, with local result limiting, cancellation, and business-status-aware retries. SerpBase defaults to a 120-second overall deadline; explicit request/configuration deadlines take precedence and other providers are unchanged. Use direct HTTP with no additional SDK dependency.

💬 Review

• 471 tests pass locally, including fake-timer checks of cancellation deadlines and timeout precedence. Wire tests cover all six endpoints; Pi tests verify only web_search is registered and follow a feature ID from model-visible Maps output into Maps Detail.
• PR CI is green on Node.js 22, 24, and 26, including formatting and the installed CLI/Pi package smoke test.
• Manual CI with a live Maps search passed on commit 8f3a302, using the environment-scoped SERPBASE_API_KEY. Latest local news/maps retries with the longer default returned upstream 1503: service unavailable, rather than a client timeout.
• Real local API checks passed for every endpoint, including Maps → Maps Detail, plus desktop/mobile and page-2 organic results. Initial detail timeout and concurrent-call throttling were observed; a sequential detail retry succeeded.
• Live responses exposed a Google navigation-logo artifact and ambiguous video time fields. Narrow regression-tested handling excludes the navigation asset and avoids inventing publication dates or durations.
• Empty result lists still have no per-result context; upstream data quality, latency, and account limits remain provider-dependent.

_{✅ Closes #45}