Fix chat retrieval payload guard

[suguanYang]

Summary

• Bound model-facing merged retrieval evidence by namespace/topK and cap long retrieval text before it is passed back into Gemini.
• Return only newly retrieved chunks/assets from each harness retrieve tool result instead of replaying the full accumulated evidence ledger.
• Cap Knowhere response log arrays while preserving raw result/reference counts.

Verification

• pnpm exec vitest run src/agent-harness/runtime.test.ts
• pnpm exec vitest run src/domains/chat/index.test.ts
• pnpm run typecheck
• pnpm run lint
• pnpm test

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
knowhere-notebook-staging	Ready	Preview, Comment	Jul 5, 2026 7:10pm

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
knowhere-notebook	Ignored		Jul 5, 2026 7:10pm