docs: rewrite README with verified eval results and architecture flow

[MomoSolaris8]

docs: lead README with verified evaluation results

Restructure the README so the first screen shows measured evidence
instead of prose.

• Add a results-first opening: one-line positioning, design-decision
  table with file references, and the real claim-review eval output
  (12/12 cases, 4 gated metrics, dataset fingerprint)
• Replace the mermaid architecture graph with a plain-text request
  path from api/ask.py through the LangGraph nodes to AskResponse
• Document the three external gateways (LLM registry, MCP tools,
  Postgres/pgvector) and why the graph is deliberately static
• Fix the online benchmark command: it is
  supportagent.evaluation.online_runner, not a --suite flag
• Translate the end-to-end test section to English and demote its
  stray H1 heading to H3

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
support-agent-one-coyl	Ready	Preview	Aug 8, 2026 2:52pm