Refactor to a Gemini browser-first runtime

[Druvith]

Summary

• refactor the project into a Gemini-only, browser-first CLI runtime built on Playwright and Rich
• remove the old OCR, desktop-control, multi-provider, API, and Docker runtime paths
• add deterministic pytest coverage, local fixture pages, and a live smoke gate for browser validation

What changed

• replace the old runtime loop with focused cli, config, gemini_browser_agent, artifacts, and typed model modules
• switch packaging to pyproject.toml with a smaller dependency set and update the package publish workflow
• add structured run artifacts, simplified Gemini error messages, primary-to-gemini-flash-lite-latest planner fallback, and keep headed operate run sessions open until Enter is pressed
• rewrite the README / AGENT / CONTRIBUTING docs to match the new scope and usage
• clean up repo clutter and extend .gitignore for local prompt experiments and editor metadata

Validation

• pytest -q
• python -m operate.main doctor
• python -m build