feat: add Tavily as search engine option in research_web

[ParakhJaggi]

Summary

• Added Tavily as a configurable search engine alongside existing providers (duckduckgo, bing, searxng, serper)
• Implemented _search_tavily() helper using TavilyClient from tavily-python, with proper error handling for missing API keys, import errors, and API failures
• Added tavily_api_key parameter to SearchConfig model and search_on_web() function
• Added 7 new test cases covering: config validation, successful search, missing/empty API key, API errors, empty results, and end-to-end dispatch

Files changed

• scrapegraphai/utils/research_web.py — Added tavily to valid engines, tavily_api_key field, _search_tavily() function, and tavily branch in search_on_web()
• tests/utils/research_web_test.py — Added 7 tavily test cases using mocked TavilyClient
• pyproject.toml — Added tavily-python as optional dependency under [project.optional-dependencies]

Dependency changes

• Added tavily-python>=0.5.0 as optional dependency (pip install scrapegraphai[tavily])

Environment variable changes

• TAVILY_API_KEY — Required when using search_engine="tavily", passed via tavily_api_key parameter

Notes for reviewers

• Tavily is added as an additive option — existing search engines are unchanged
• tavily-python is imported lazily inside _search_tavily() to avoid hard dependency
• All 7 new tests pass with mocked TavilyClient (no API key needed for tests)

🤖 Generated with Claude Code

Automated Review

• Passed after 1 attempt(s)
• Final review: The Tavily migration is correct and complete. The implementation follows existing codebase patterns precisely: lazy import, consistent error handling with SearchConfigError/SearchRequestError, URL extraction from response["results"], and optional dependency in pyproject.toml. The 7 new tests are well-structured with proper sys.modules mocking. Three minor issues were found (type annotation, missing docstring entry, weak integration test assertion) but none block approval.