feat: add Tavily as configurable search provider alongside Google PSE

[tavily-integrations]

Summary

• Added Tavily as a configurable parallel/fallback search provider in scripts/search.py
• New --provider flag accepts google (default), tavily, or auto (Google PSE with Tavily fallback on 403/429)
• Tavily results are normalized into the same Markdown output format as Google PSE for consistent caller experience
• Tavily is fully optional — the script continues to work without TAVILY_API_KEY set

Files changed

• scripts/search.py — Added get_tavily_client(), tavily_search(), format_tavily_results() functions; added --provider CLI argument; updated search() to route by provider and fall back to Tavily on quota errors
• SKILL.md — Updated description, search tool priority table, usage examples, options table, and error handling table to document Tavily provider
• requirements.txt — New file listing requests and tavily-python dependencies

Dependency changes

• Added tavily-python to requirements.txt

Environment variable changes

• Added optional TAVILY_API_KEY — enables --provider tavily and --provider auto fallback

Notes for reviewers

• This is an additive change: existing Google PSE behavior is fully preserved when --provider google (the default) is used
• tavily-python is imported lazily so the script doesn't fail if the package is not installed and Tavily is not being used
• The --provider auto mode tries Google PSE first, then falls back to Tavily only on HTTP 403/429 (quota/rate-limit errors)

Automated Review

• Passed after 2 attempt(s)
• Final review: The implementation correctly addresses all three prior review issues: unsupported-flag stderr warnings with date mapping, elimination of redundant TavilyClient instantiation in the auto fallback path, and pinned dependency versions in requirements.txt. Tavily SDK usage (TavilyClient, search(), time_range values, include_domains) is correct. Documentation in SKILL.md is accurate and complete. Two minor issues are noted but do not block approval.