docs: add PR1 agent harness MVP implementation plan

[shihao-Liiiii]

Summary

• Add docs/PR1-Agent-Harness-MVP实施方案.md as the detailed implementation plan for the first Agent Harness code PR.
• Scope PR 1 to the harness foundation only: dependencies, AgentDefinition, AgentRun/Event/Artifact persistence, workspace, ToolRegistry, PSOP gateway ChatModel adapter, DeepAgents-based runner, demo agent, and Agent API.
• Define the demo acceptance path: psop-demo must run through DeepAgents, call at least one tool, persist events/artifacts, and be queryable through API.
• Clarify that fake models are only for deterministic automated tests; manual PR acceptance must run the demo agent with the real qwen3.7-plus text model through PSOP LlmInferenceGateway.

Validation

Documentation-only change. No tests run.