feat: Phase 3 - parallel execution, progress reporting, result caching + AgentChat bug fixes

Phase 3 能力:
- DAG 并行执行 (workflow_engine): asyncio.gather 并行执行就绪节点
- Debate 并行 (orchestrator): for 循环改为 asyncio.gather
- 粒度进度上报 (workflow_engine + tasks + websocket): Redis 推送 + DB 降级
- 工具结果缓存 (tool_manager): 确定性工具默认开启缓存
- LLM 响应缓存 (core): messages[-4:] + model 哈希，5min TTL

AgentChat bug 修复 (Gitea #1-#5):
- #1 SSE 降级重复空消息: fallback POST 前移除占位消息
- #2 streamTimeout 泄漏: while 正常退出后 clearTimeout
- #3 loading 闪烁: final/error 事件中提前设 loading=false
- #4 SSE 事件类型对齐: 确认匹配，未知类型加 console.warn
- #5 retryMessage 流式残留: 重试时清理占位消息

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

backend/app/agent_runtime/orchestrator.py

@@ -9,6 +9,7 @@ Agent Orchestrator — 多 Agent 编排引擎。
 """
 from __future__ import annotations
 
+import asyncio
 import json
 import logging
 import uuid
@@ -347,9 +348,10 @@ class AgentOrchestrator:
         steps: List[OrchestratorStep] = []
         agent_outputs: List[Dict[str, Any]] = []
 
-        # 第一阶段：所有 Agent 独立回答
+        # 第一阶段：所有 Agent 并行独立回答
+        runtimes = []
         for agent_cfg in agents:
-            runtime = AgentRuntime(
+            runtimes.append(AgentRuntime(
                 AgentConfig(
                     name=agent_cfg.name,
                     system_prompt=agent_cfg.system_prompt,
@@ -364,8 +366,30 @@ class AgentOrchestrator:
                     ),
                 ),
                 on_llm_call=on_llm_call,
-            )
-            result = await runtime.run(question)
+            ))
+
+        results = await asyncio.gather(
+            *[rt.run(question) for rt in runtimes],
+            return_exceptions=True,
+        )
+
+        for i, agent_cfg in enumerate(agents):
+            result = results[i]
+            if isinstance(result, BaseException):
+                step = OrchestratorStep(
+                    agent_id=agent_cfg.id,
+                    agent_name=agent_cfg.name,
+                    input=question,
+                    output="",
+                    error=str(result),
+                )
+                steps.append(step)
+                agent_outputs.append({
+                    "agent_id": agent_cfg.id,
+                    "agent_name": agent_cfg.name,
+                    "output": f"[错误] {result}",
+                })
+                continue
 
             step = OrchestratorStep(
                 agent_id=agent_cfg.id,