feat(pi-permission-system): AutoReview 改为跟随主模型的对话式工具调用审查

AutoReview：

- 审查器回到 Codex Guardian 风格的对话式审查，默认跟随会话当前主模型
  （每次审查实时读取 ctx.model）；provider/model 必须同时设置才固定审查
  模型，只设其一视为配置错误并转人工。
- 结论只通过 approve / reject 两个工具给出，参数仅 reason；不强制
  toolChoice，正文文本（即使是合法 JSON）一律忽略。
- approve 自动放行；reject 转人工并在弹窗显示理由（标注模型生成）。
  超时、Provider 错误、截断、未调用工具、多次调用、参数无效等异常同样
  转人工，并附固定的系统说明；用户按 Esc 中止时直接拒绝。
- 模型调用改为 modelRegistry.streamSimple(...).result()：Pi 1.3.0 面向
  扩展的 ModelRegistry 没有 completeSimple，原写法在真实环境中每次都会
  转人工。
- 对话记录保留可信度标注；工具结果在有更短的 pi-compaction 摘要时替换
  为摘要；token 估算改为非 ASCII 字符按 1 token 计；预算可配置。
- 同一条用户消息内完全相同的请求只缓存 approve 结论。
- 默认 reasoning low、总超时 30 秒、最多 2 次尝试、maxOutputTokens 4000。

删除：

- 未发布过的 TypeSafe Jev 分类器实现、DenialCircuitBreaker。
- 整个 annotator 机制（model-explanation、注册表、registerAnnotator、
  provider-timeout）；annotators 配置键保留为已弃用并忽略，避免旧配置
  被拒绝。

测试：

- 新增 reviewer / transcript / 弹窗说明 / 默认配置测试；清理重复、恒真
  或只测实现细节的旧测试。
- 使用 deepseek-v4.1-flash 实测 5 个场景各 5 次，25 次结论全部符合预期。

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

pi-permission-system/test/hard-guard-loader.test.ts

@@ -49,4 +49,81 @@ describe("HardGuardPolicyLoader", () => {
     expect(loader.loadGlobalConfig().permission).toEqual({ "*": "deny" });
     expect(loader.getConfigIssues().join("\n")).toContain("allow or deny");
   });
+
+  it("rejects an unsupported version instead of interpreting it", () => {
+    const root = mkdtempSync(join(tmpdir(), "hard-guard-loader-"));
+    roots.push(root);
+    const userPath = join(root, "hard-guards.jsonc");
+    writeFileSync(
+      userPath,
+      `{ "version": 2, "permission": { "*": "allow" } }`,
+    );
+
+    const loader = new HardGuardPolicyLoader(
+      userPath,
+      join(root, "missing-fallback.jsonc"),
+    );
+
+    expect(loader.loadGlobalConfig().permission).toEqual({ "*": "deny" });
+    expect(loader.getConfigIssues().join("\n")).toContain(
+      "unsupported hard guard version",
+    );
+  });
+
+  it("keeps the last good snapshot when the file turns invalid", () => {
+    const root = mkdtempSync(join(tmpdir(), "hard-guard-loader-"));
+    roots.push(root);
+    const userPath = join(root, "hard-guards.jsonc");
+    writeFileSync(
+      userPath,
+      `{ "version": 1, "permission": { "*": "allow", "~/.ssh/**": "deny" } }`,
+    );
+
+    const loader = new HardGuardPolicyLoader(
+      userPath,
+      join(root, "missing-fallback.jsonc"),
+    );
+    const first = loader.loadGlobalConfig();
+    expect(first.permission).toEqual({
+      "*": "allow",
+      "~/.ssh/**": "deny",
+    });
+
+    // A corrupt edit must not weaken the boundary: the last valid snapshot
+    // keeps applying while the file stays broken.
+    writeFileSync(userPath, "{ not json");
+    const second = loader.loadGlobalConfig();
+
+    expect(second).toEqual(first);
+    expect(loader.getConfigIssues().join("\n")).toContain("Failed to load");
+  });
+
+  it("restores the file's own rules once it becomes valid again", () => {
+    const root = mkdtempSync(join(tmpdir(), "hard-guard-loader-"));
+    roots.push(root);
+    const userPath = join(root, "hard-guards.jsonc");
+    writeFileSync(
+      userPath,
+      `{ "version": 1, "permission": { "*": "allow", "~/.ssh/**": "deny" } }`,
+    );
+
+    const loader = new HardGuardPolicyLoader(
+      userPath,
+      join(root, "missing-fallback.jsonc"),
+    );
+    loader.loadGlobalConfig();
+    writeFileSync(userPath, "{ not json");
+    loader.loadGlobalConfig();
+
+    writeFileSync(
+      userPath,
+      `{ "version": 1, "permission": { "*": "allow", "~/.aws/**": "deny" } }`,
+    );
+    const restored = loader.loadGlobalConfig();
+
+    expect(restored.permission).toEqual({
+      "*": "allow",
+      "~/.aws/**": "deny",
+    });
+  });
 });