Replace temperature with reasoning effort and retry in-stream errors

- Providers get a default reasoning effort (low..max, empty = not sent);
  chat can override it per message. OpenAI sends reasoning_effort,
  Claude sends output_config.effort with adaptive thinking, and thinking
  blocks are replayed during the tool loop.
- Show a thinking indicator while waiting for model output.
- Surface in-stream upstream errors (e.g. overloaded) instead of
  breaking the SSE stream, and retry them up to 3 times when nothing
  has been emitted yet.

web/src/pages/chat/components/MessageList.vue

@@ -18,6 +18,8 @@ export interface DisplayItem {
   error?: string
   /** 生成中：这次模型调用已结束，后续文字另起一条 */
   closed?: boolean
+  /** 生成中：模型正在思考，还没有输出文字 */
+  thinking?: boolean
 }
 
 defineProps<{ items: DisplayItem[] }>()
@@ -53,6 +55,9 @@ function prettyResult(result: string) {
       </div>
       <div v-else class="flex flex-col gap-2">
         <!-- markdown-it 已关闭 html，模型输出中的标签按文本显示 -->
+        <div v-if="item.thinking" class="text-xs text-[var(--el-text-color-secondary)]">
+          思考中…
+        </div>
         <div v-if="item.text" class="markdown-body" v-html="renderMarkdown(item.text)" />
         <el-collapse v-if="item.toolCalls.length" class="tool-cards">
           <el-collapse-item v-for="call in item.toolCalls" :key="call.id" :name="call.id">