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>

frontend/src/views/AgentChat.vue

@@ -439,6 +439,7 @@ async function sendMessage() {
 
       // 尝试 SSE 流式（带超时控制）
       let usedStreaming = false
+      let placeholderIdx = -1
       streamingActive.value = false
       const abortController = new AbortController()
       const streamTimeout = setTimeout(() => abortController.abort(), 60000)
@@ -462,8 +463,8 @@ async function sendMessage() {
             role: 'assistant', content: '', timestamp: Date.now(),
             steps: [], _traceOpen: true, iterations: 0, tool_calls_made: 0,
           }
-          const idx = messages.value[key].push(msg) - 1
-          const currentMsg = messages.value[key][idx]
+          placeholderIdx = messages.value[key].push(msg) - 1
+          const currentMsg = messages.value[key][placeholderIdx]
 
           const reader = resp.body.getReader()
           const decoder = new TextDecoder()
@@ -532,14 +533,19 @@ async function sendMessage() {
                     sessionId.value[key] = data.session_id
                   }
                   streamingActive.value = false
+                  loading.value = false
                 } else if (eventType === 'error') {
                   currentMsg.content = data.content || ''
                   currentMsg.status = 'error'
                   streamingActive.value = false
+                  loading.value = false
+                } else {
+                  console.warn('[AgentChat] 未知 SSE 事件类型:', eventType, data)
                 }
               } catch { /* 跳过畸形事件 */ }
             }
           }
+          clearTimeout(streamTimeout)
         }
       } catch {
         clearTimeout(streamTimeout)
@@ -549,6 +555,11 @@ async function sendMessage() {
       }
 
       if (!usedStreaming) {
+        // 移除 SSE 阶段残留的占位消息（Issue #1）
+        if (placeholderIdx >= 0 && placeholderIdx < messages.value[key].length) {
+          messages.value[key].splice(placeholderIdx, 1)
+        }
+
         // 降级：标准 POST 请求
         const fallbackEndpoint = currentAgentId.value
           ? `/api/v1/agent-chat/${currentAgentId.value}`
@@ -613,9 +624,11 @@ function retryMessage(idx: number) {
 
   // 查找错误消息之前的最后一条用户消息
   let userMsg = ''
+  let userIdx = -1
   for (let i = idx - 1; i >= 0; i--) {
     if (msgs[i].role === 'user') {
       userMsg = msgs[i].content
+      userIdx = i
       break
     }
   }
@@ -624,15 +637,15 @@ function retryMessage(idx: number) {
     return
   }
 
-  // 移除该错误消息及关联的用户消息
-  const removeIndices: number[] = []
-  for (let i = idx - 1; i >= 0; i--) {
-    if (msgs[i].role === 'user' && msgs[i].content === userMsg) {
+  // 收集需要移除的消息：用户消息 + 错误消息 + 中间的流式占位消息
+  const removeIndices: number[] = [userIdx, idx]
+  for (let i = userIdx + 1; i < idx; i++) {
+    const m = msgs[i]
+    // 流式占位消息：assistant 角色，内容为空或有 steps 但无实质内容
+    if (m.role === 'assistant' && (!m.content || m.content === '') && (m.steps?.length || 0) > 0) {
       removeIndices.push(i)
-      break
     }
   }
-  removeIndices.push(idx)
 
   // 从后往前删除，避免 index 错乱
   removeIndices.sort((a, b) => b - a)