feat: 新增模型管理功能（供应商 + 模型 CRUD）

- 新增 providers/models 数据库表、迁移和数据访问层
- 新增 15 个后端 API 路由（供应商/模型 CRUD + 连通性测试）
- 新增 AI 服务层（registry.ts: buildProviderRegistry + testProviderConnection）
- 新增前端模型管理页面（Tabs: 供应商/模型，含表格、表单、工具栏）
- 新增前端 hooks（use-providers, use-models）
- 新增共享类型和 MODEL_CAPABILITIES 常量
- 新增 10 个测试文件（66 个测试用例，4 个因 bun test ESM 兼容问题待修复）
- 更新开发文档（architecture, backend, frontend）
- 附带 apply-review 修复：统一错误响应、提取共享常量、清理重复测试

注意：registry.test.ts 中 4 个测试因 bun test 无法解析
createProviderRegistry ESM 导出而失败，详情见 context.md

tests/web/hooks/use-models.test.ts

@@ -0,0 +1,105 @@
+import { describe, expect, test } from "bun:test";
+
+import {
+  createModel,
+  deleteModel,
+  disableModel,
+  enableModel,
+  fetchModel,
+  fetchModelList,
+  updateModel,
+} from "../../../src/web/hooks/use-models";
+import { installFetchMock, jsonResponse } from "../test-utils";
+
+const MODEL = {
+  capabilities: ["text"] as Array<"text">,
+  contextLength: null,
+  createdAt: "2024-01-01T00:00:00.000Z",
+  enabled: true,
+  id: "m1",
+  maxOutputTokens: null,
+  modelId: "gpt-4o",
+  name: "GPT-4o",
+  providerId: "pv1",
+  updatedAt: "2024-01-01T00:00:00.000Z",
+};
+
+async function expectRejectsWithMessage(action: () => Promise<unknown>, message: string) {
+  try {
+    await action();
+    throw new Error("expected rejection");
+  } catch (error) {
+    expect(error).toBeInstanceOf(Error);
+    expect((error as Error).message).toBe(message);
+  }
+}
+
+function jsonBody(body: BodyInit | null | undefined): unknown {
+  return JSON.parse(typeof body === "string" ? body : "{}");
+}
+
+describe("use-models request helpers", () => {
+  test("fetchModelList 按协议拼接 query 参数（含 providerId）", async () => {
+    const calls = installFetchMock(() => jsonResponse({ items: [MODEL], page: 1, pageSize: 20, total: 1 }));
+
+    const result = await fetchModelList({ keyword: "GPT", page: 1, pageSize: 20, providerId: "pv1" });
+
+    expect(result.items).toHaveLength(1);
+    expect(calls[0]?.method).toBe("GET");
+    expect(calls[0]?.url).toContain("providerId=pv1");
+    expect(calls[0]?.url).toContain("keyword=GPT");
+  });
+
+  test("模型 CRUD 与 enable/disable 使用正确 method、URL 与 body", async () => {
+    const calls = installFetchMock((call) => {
+      if (call.method === "DELETE") return new Response(null, { status: 204 });
+      return jsonResponse(
+        { model: MODEL },
+        { status: call.method === "POST" && call.url === "/api/models" ? 201 : 200 },
+      );
+    });
+
+    await createModel({
+      capabilities: ["text"],
+      modelId: "gpt-4o",
+      name: "GPT-4o",
+      providerId: "pv1",
+    });
+    await updateModel("m1", { name: "GPT-4o Mini" });
+    await enableModel("m1");
+    await disableModel("m1");
+    await deleteModel("m1");
+    await fetchModel("m1");
+
+    expect(calls.map((call) => `${call.method} ${call.url}`)).toEqual([
+      "POST /api/models",
+      "PATCH /api/models/m1",
+      "POST /api/models/m1/enable",
+      "POST /api/models/m1/disable",
+      "DELETE /api/models/m1",
+      "GET /api/models/m1",
+    ]);
+    expect(jsonBody(calls[0]?.body)).toEqual({
+      capabilities: ["text"],
+      modelId: "gpt-4o",
+      name: "GPT-4o",
+      providerId: "pv1",
+    });
+    expect(jsonBody(calls[1]?.body)).toEqual({ name: "GPT-4o Mini" });
+  });
+
+  test("错误响应优先使用后端 error 字段", async () => {
+    installFetchMock(() => jsonResponse({ error: "模型名称已存在" }, { status: 409 }));
+
+    await expectRejectsWithMessage(
+      () => createModel({ capabilities: ["text"], modelId: "gpt-4o", name: "重复", providerId: "pv1" }),
+      "模型名称已存在",
+    );
+  });
+
+  test("非 JSON 错误响应回退到 HTTP 状态", async () => {
+    installFetchMock(() => new Response("broken", { status: 500 }));
+
+    await expectRejectsWithMessage(() => fetchModel("m-missing"), "HTTP 500");
+  });
+});