feat: migrate search handler to support Tavily alongside DuckDuckGo

[tavily-integrations]

Summary

• Added tavily-python dependency alongside existing duckduckgo-search in requirements.txt
• Implemented tavily_web_search() in server-fetch.py that calls TavilyClient.search() and normalizes results to the same dict shape as DuckDuckGo results
• Added _get_search_provider() helper that reads SEARCH_PROVIDER env var; defaults to tavily when TAVILY_API_KEY is set, otherwise falls back to DuckDuckGo
• Updated Fetcher.web_search to dispatch to Tavily or DuckDuckGo based on configured provider
• Added TAVILY_API_KEY and SEARCH_PROVIDER to .env.example with documentation
• Added __pycache__/ and *.pyc to .gitignore

Files changed

• requirements.txt — added tavily-python==0.5.0
• backend/tools/server-fetch.py — Tavily import, search function, provider dispatch
• .env.example — new env var documentation
• .gitignore — ignore Python bytecode artifacts

Dependency changes

• Added tavily-python==0.5.0 to requirements.txt

Env var changes

• Added TAVILY_API_KEY (required when using Tavily provider)
• Added SEARCH_PROVIDER (optional; values: tavily | duckduckgo)

Notes for reviewers

• DuckDuckGo code path is fully preserved — no existing functionality is removed
• Tavily import is soft (try/except) matching the existing DDGS import pattern
• Provider auto-selects Tavily when TAVILY_API_KEY is present; explicit SEARCH_PROVIDER overrides

🤖 Generated with Claude Code

Automated Review

• Passed after 1 attempt(s)
• Final review: The migration is correct and complete. Tavily is added as an additive, configurable parallel provider alongside DuckDuckGo with no regressions. The soft-import pattern, env var dispatch logic, result normalization, and dependency additions are all implemented correctly. Four minor issues are noted but none block approval.