feat: 重构配置生命周期为 Authoring/Normalized/Resolved 三层

将变量替换和 expect 简写展开统一放入 Normalized 阶段，
运行时 AJV 使用 Normalized schema，导出 schema 面向 Authoring Config。

主要变更：
- 新增 normalizer.ts 实现 normalizeAuthoringConfig()
- 拆分 Authoring/Normalized 双 schema，checker 接口支持 authoring/normalized 片段
- config-loader 流程：normalize → Normalized AJV → semantic → resolve
- validator 兼容层自动分派 raw/normalized expect 形态
- 删除 rawExpect，store.expect 列写入 null
- Authoring schema 对 integer/boolean/enum 字段接受变量引用
- 修复 DB/HTTP validate 入口守卫和 LLM options integer 变量引用
- 优化 compact() 避免 undefined 覆盖隐患
- 移除 content.ts 恒为 true 的前置条件
- 同步 5 个主规范并归档 change

tests/server/checker/runner/llm/schema-validate-resolve.test.ts

@@ -60,9 +60,11 @@ describe("LlmChecker schema", () => {
     expect(checker?.configKey).toBe("llm");
   });
 
-  test("schemas 包含 config、expect", () => {
+  test("schemas 包含 authoring 和 normalized config/expect", () => {
     expect(checker).toBeDefined();
-    expect(Object.keys(checker!.schemas).sort()).toEqual(["config", "expect"].sort());
+    expect(Object.keys(checker!.schemas).sort()).toEqual(["authoring", "normalized"].sort());
+    expect(checker!.schemas.authoring.config).toBeDefined();
+    expect(checker!.schemas.normalized.expect).toBeDefined();
   });
 });
 
@@ -249,13 +251,13 @@ describe("LlmChecker resolve", () => {
     expect(resolved.group).toBe("default");
     expect(resolved.intervalMs).toBe(30000);
     expect(resolved.timeoutMs).toBe(10000);
-    expect(resolved.rawExpect).toBeUndefined();
+    expect("rawExpect" in resolved).toBe(false);
     expect(resolved.expect).toEqual({ status: [200] });
   });
 
   test("stream mode 未配置 expect.stream 时不物化 completed", () => {
     const raw = makeRawTarget({
-      expect: { output: [{ contains: "OK" }] },
+      expect: { output: [{ kind: "value", matcher: { contains: "OK" } }] },
       llm: {
         mode: "stream",
         model: "gpt-4o-mini",
@@ -267,13 +269,13 @@ describe("LlmChecker resolve", () => {
 
     const resolved = asLlm(checker.resolve(raw, makeResolveContext()));
 
-    expect(resolved.rawExpect).toEqual({ output: [{ contains: "OK" }] });
+    expect("rawExpect" in resolved).toBe(false);
     expect(resolved.expect?.stream).toBeUndefined();
   });
 
   test("配置 expect.stream 但省略 completed 时默认 true", () => {
     const raw = makeRawTarget({
-      expect: { stream: { firstTokenMs: 100 } },
+      expect: { stream: { firstTokenMs: { equals: 100 } } },
       llm: {
         mode: "stream",
         model: "gpt-4o-mini",
@@ -285,7 +287,7 @@ describe("LlmChecker resolve", () => {
 
     const resolved = asLlm(checker.resolve(raw, makeResolveContext()));
 
-    expect(resolved.rawExpect).toEqual({ stream: { firstTokenMs: 100 } });
+    expect("rawExpect" in resolved).toBe(false);
     expect(resolved.expect?.stream).toEqual({ completed: true, firstTokenMs: { equals: 100 } });
   });