Fix multi-agent demo, add tests, docs site, CI, and SEO

[rrahimi-uci]

Repairs the latent bugs that prevented the orchestrator demo from ever working, adds a real test suite + packaging, rewrites the docs, and ships a search-engine-optimized GitHub Pages site.

Correctness fixes

• Doubled backslashes across agent/example sources rendered as literal \n in output and broke every regex (number/data extraction crashed). Collapsed to single backslashes.
• base_agent: serialize with model_dump(mode="json") so datetime survives the JSON-RPC hop over aiohttp; added /health and /.well-known/agent.json.
• Client error detection: a JSON-RPC response always carries an "error" key (null on success), so the old if "error" in result check misread every success as an error — the demo only ever printed Agent returned error: None. Fixed in A2AClient, BaseAgent.send_message_to_agent, and simple_a2a_client.
• Removed dead, un-importable root __init__.py.

Tooling & tests

• 34-test pytest suite (in-process TestClient + one live-socket integration test guarding the JSON-RPC parsing regression).
• pyproject.toml (packaging + pytest config), requirements-dev.txt.
• Startup scripts the README/tests referenced but that never existed: start_math_agent, start_data_agent, run_demo, launch_system; moved the live smoke test to scripts/smoke_test.py.

Docs & site

• Rewrote README; added ARCHITECTURE.md with mermaid diagrams.
• docs/index.html landing page with full SEO (meta, Open Graph, Twitter cards, canonical, JSON-LD), generated og-image.png, sitemap.xml, robots.txt, .nojekyll; SEO tags on the architecture page.
• CI workflow (pytest, Py 3.9–3.12) and Pages deploy workflow; MIT LICENSE.

Verification

• pytest → 34 passed
• python scripts/launch_system.py → real multi-agent collaboration producing correct stats + visualization output.

🤖 Generated with Claude Code