feat: Phase 4 - LLM/Agent fallback chain, cross-agent knowledge sharing, async agent execution

- 4.1 Fallback chain: LLM fallback_llm config in AgentLLMConfig, retry with alternate model on API failure; Agent fallback_agent in DAG nodes
- 4.2 Knowledge sharing: GlobalKnowledge model with embedding-based semantic search, auto-extraction of tool names as tags after execution
- 4.3 Async execution: execute_agent_task fully implemented with AgentRuntime, scheduler dual-path for workflow/non-workflow agents

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/app/agent_runtime/schemas.py

@@ -46,6 +46,7 @@ class AgentLLMConfig(BaseModel):
     self_review_threshold: float = 0.6  # self-review 通过阈值（0-1）
     cache_enabled: bool = False          # LLM 响应缓存（默认关闭，语义缓存有风险）
     cache_ttl_ms: int = 300000           # LLM 缓存 TTL，默认 5 分钟
+    fallback_llm: Optional[Dict[str, Any]] = None  # 降级模型配置 {provider, model, api_key, base_url}
 
 
 class AgentBudgetConfig(BaseModel):