W17: extract ai_common shared module + fix anthropic data race + brace bugs

- New plugins_upper/ai_common/ static library: shared PluginConfig, ToolCallAccum,
  StreamContext, secure_zero, extract_host_port, serialize_tool_calls, free_chat_result
- Refactored openai/anthropic plugins to use dstalk_ai:: namespace from ai_common
- Fixed anthropic g_config raw pointer → std::atomic (data race)
- Added SSE parse error counter with threshold abort (kMaxSseParseErrors=5)
- Fixed missing closing brace in both plugins' error-body catch block
- Updated test targets: ai_common include path + link, using namespace dstalk_ai
- plugin_loader_test: added stub_unreg + service_registry.cpp for unregister_service
- Includes pre-existing uncommitted changes from prior waves

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

docs/tutorial/quick-start.md

@@ -6,35 +6,65 @@
 
 ## 1. 安装工具链
 
-dstalk 需要 CMake、Ninja、LLVM/Clang 和 Conan2。`setup.bat` 全自动下载安装到 `tools/` 目录。
+dstalk 需要 CMake、Ninja、LLVM/Clang 和 Conan2。`tools/` 下的脚本全自动安装。
 
+**Windows:**
 ```bash
 cd tools
 setup.bat
 ```
 
+**Mac:**
+```bash
+# 方式 A: 手动安装后运行脚本
+brew install cmake ninja
+cd tools && bash setup.sh
+
+# 方式 B: 脚本全自动 (推荐)
+cd tools && bash setup.sh
+```
+
+**Linux (Ubuntu/Debian):**
+```bash
+# 方式 A: 手动安装后运行脚本
+sudo apt install cmake ninja-build
+cd tools && bash setup.sh
+
+# 方式 B: 脚本全自动 (推荐)
+cd tools && bash setup.sh
+```
+
 > **前提**: 系统需已安装 Python 3.10+。
 >
 > 网络不畅时, 可手动下载放入对应目录:
-> - [Ninja](https://github.com/ninja-build/ninja/releases) → `tools/ninja/ninja.exe`
-> - [CMake](https://cmake.org/download/) → `tools/cmake/bin/cmake.exe`
-> - [LLVM](https://github.com/llvm/llvm-project/releases) → `tools/llvm/bin/clang.exe`
-> - Conan2 通过 pip 安装到 `tools/.venv/Scripts/conan.exe`
+> - [Ninja](https://github.com/ninja-build/ninja/releases) → `tools/ninja/ninja` (或 `ninja.exe`)
+> - [CMake](https://cmake.org/download/) → `tools/cmake/bin/cmake` (或 `cmake.exe`)
+> - [LLVM](https://github.com/llvm/llvm-project/releases) → `tools/llvm/bin/clang` (或 `clang.exe`)
+> - Conan2 通过 pip 安装到 `tools/.venv/Scripts/conan.exe` (Win) 或 `tools/.venv/bin/conan` (Mac/Linux)
 
 ---
 
 ## 2. 编译
 
-项目根目录提供 `build.bat`, 一键完成: Conan 拉取依赖 -> CMake 配置 -> Ninja 编译。
+项目根目录提供 `build.bat` (Windows) / `build.sh` (Mac/Linux), 一键完成: Conan 拉取依赖 -> CMake 配置 -> Ninja 编译。
 
+**Windows:**
 ```bash
 build.bat
 ```
 
-编译产物输出到 `build/` 目录。核心产物:
-- `build/dstalk_core/dstalk.dll` —— 核心 DLL
-- `build/dstalk_cli/dstalk_cli.exe` —— 命令行前端
-- `build/plugins/*.dll` —— 功能插件
+**Mac / Linux:**
+```bash
+bash build.sh
+```
+
+编译产物输出到 `build/` 目录:
+
+| 产物 | Windows | Mac | Linux |
+|------|---------|-----|-------|
+| 核心 DLL | `build/bin/dstalk.dll` | `build/bin/libdstalk.dylib` | `build/bin/libdstalk.so` |
+| CLI 前端 | `build/bin/dstalk_cli.exe` | `build/bin/dstalk_cli` | `build/bin/dstalk_cli` |
+| 功能插件 | `build/plugins/*.dll` | `build/plugins/*.dylib` | `build/plugins/*.so` |
 
 ---
 
@@ -49,15 +79,15 @@ build.bat
 **config.toml 示例:**
 
 ```toml
-# 选择 AI 后端插件: ai.openai 或 ai.anthropic
-ai.provider = "ai.openai"
+# 选择 AI 后端插件: ai_openai 或 ai_anthropic
+ai.provider = "ai_openai"
 
 # OpenAI-compatible
 api.base_url = "https://api.openai.com/v1"
 api.api_key = "sk-xxxxxxxx"
 api.model = "gpt-4o"
 
-# Anthropic Claude (切换 ai.provider 为 "ai.anthropic" 即可)
+# Anthropic Claude (切换 ai.provider 为 "ai_anthropic" 即可)
 # api.base_url = "https://api.anthropic.com/v1"
 # api.api_key = "sk-ant-xxxxxxxx"
 # api.model = "claude-opus-4-20250514"
@@ -71,8 +101,14 @@ api.model = "gpt-4o"
 
 ## 4. 运行 dstalk_cli
 
+**Windows:**
 ```bash
-build/dstalk_cli/dstalk_cli.exe
+build\bin\dstalk_cli.exe
+```
+
+**Mac / Linux:**
+```bash
+build/bin/dstalk_cli
 ```
 
 启动后显示欢迎横幅: