fix: 修复 Agent 流式对话无响应和工具 schema 兼容性问题

- 在 `run_stream()` LLM 调用前 yield `think` 事件，前端即时显示"思考中..."
- 修复 tool schema 规范化逻辑：`{"function":{...}}` 格式缺少 `type` 字段导致 LLM API 拒绝
- 启动时从数据库加载自定义工具（`load_tools_from_db`），解决重启后工具丢失
- 前端 SSE 添加 60s 超时保护，任何事件类型均触发 `receivedFirstEvent`
- 流式失败自动降级到非流式 POST
- 添加 `scripts/seed_coding_agent.py` 和 `scripts/test_coding_agent.py`

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

backend/app/agent_runtime/core.py

@@ -83,8 +83,9 @@ class AgentRuntime:
             system_prompt=self.config.system_prompt,
             user_id=self.config.user_id,
         )
+        _mem_scope = self.config.memory_scope_id or self.config.user_id or self.config.name
         self.memory = memory or AgentMemory(
-            scope_id=self.config.user_id or self.config.name,
+            scope_id=_mem_scope,
             max_history=self.config.memory.max_history_messages,
             persist=self.config.memory.persist_to_db,
         )
@@ -398,6 +399,9 @@ class AgentRuntime:
                            "truncated": True}
                     return
 
+            # think 事件：告知前端 Agent 正在思考（让 UI 即时反馈，避免假死感）
+            yield {"type": "think", "content": "", "reasoning": None, "iteration": self.context.iteration}
+
             # 调用 LLM
             try:
                 response = await llm.chat(
@@ -687,7 +691,24 @@ class _LLMClient:
         if self._config.extra_body:
             kwargs["extra_body"] = self._config.extra_body
         if tools:
-            kwargs["tools"] = tools
+            # Normalize tool schemas to OpenAI format: custom tools from the
+            # marketplace may be stored as {"name":..., "parameters":...}
+            # or {"function":{...}} without the required "type": "function".
+            normalized = []
+            for t in tools:
+                if isinstance(t, dict):
+                    if t.get("type") == "function":
+                        # Already in correct format: {"type":"function","function":{...}}
+                        normalized.append(t)
+                    elif "function" in t:
+                        # Has function key but missing type: {"function":{...}}
+                        normalized.append({"type": "function", "function": t["function"]})
+                    else:
+                        # Raw schema: {"name":..., "parameters":...}
+                        normalized.append({"type": "function", "function": t})
+                else:
+                    normalized.append(t)
+            kwargs["tools"] = normalized
             kwargs["tool_choice"] = "auto"
 
         start_time = time.perf_counter()