feat(chat): support Harness Engine clients

[sriaradhyula]

Summary

• show each agent's execution harness in agent discovery, the picker, interactive chat, and headless output
• preserve legacy server behavior by defaulting missing harness metadata to LangChain Deep Agents
• accept both camelCase and legacy snake_case tool-result identifiers
• forward portable repository, skills, and memory context unchanged
• retain Harness Engine run IDs and SSE cursors, then resume detached runs without resubmitting prompts or repeating provider/tool execution

Root cause

The CLI treated every discovered agent as a generic AG-UI Dynamic Agent and discarded additive harness metadata. It also treated an unexpected SSE EOF as successful completion and did not retain the Harness Gateway run ID or event cursor needed to reconnect to a detached AgentCore or Claude SDK run.

Compatibility

Existing Dynamic Agents and older CAIPE servers keep their current AG-UI flow. Harness metadata fields are optional, JSON additions are additive, and replay activates only when the server returns X-Harness-Run-ID.

Validation

• npm test — 32 files, 231 tests including a real-HTTP Harness Engine disconnect/replay E2E
• npm run lint
• bunx tsc --noEmit
• npm run compile

Companion server metadata change: cnoe-io/ai-platform-engineering#2401