feat: 完成全量功能开发，包括前端管理后台与后端服务优化

此提交实现了完整的知识库管理系统：
1. 新增Vue3 + Antd Vue前端管理后台，包含登录、文档管理、检索、类目设置等完整页面
2. 重构后端LLM调用抽象层，支持Ollama与OpenAI兼容服务动态切换
3. 调整默认嵌入模型配置为本地bge-m3模式
4. 优化入库任务去重逻辑与缓存清理机制
5. 完善Docker镜像构建与docker-compose部署配置
6. 修复多项测试用例与兼容性问题
7. 新增运行时配置API，支持动态调整系统参数

tests/test_runtime_settings.py

@@ -0,0 +1,254 @@
+"""RuntimeSettings 单元测试：默认值 / 加载 / 保存 / 部分更新 / 重置 / 单例
+
+每个测试通过 monkeypatch 把 RUNTIME_SETTINGS_PATH 指向独立临时文件，并在前后
+重置模块级 `_runtime_settings` 单例，避免跨测试串扰。
+"""
+
+import json
+from pathlib import Path
+
+import pytest
+
+from app.core import runtime_settings as rs
+
+
+@pytest.fixture
+def isolated_settings_path(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Path:
+    """每个测试独立持久化路径，并在前后清空模块级单例"""
+    path = tmp_path / "runtime_settings.json"
+    monkeypatch.setenv("RUNTIME_SETTINGS_PATH", str(path))
+    # 重置单例，强制下次 get_runtime_settings 重新加载
+    rs._runtime_settings = None
+    yield path
+    rs._runtime_settings = None
+
+
+# ---------------------------------------------------------------------------- #
+# 默认值
+# ---------------------------------------------------------------------------- #
+
+
+class TestDefaults:
+    def test_default_models_provider_is_ollama(self):
+        cfg = rs.RuntimeSettings()
+        assert cfg.models.summarize.provider == "ollama"
+        assert cfg.models.query.provider == "ollama"
+        assert cfg.models.classify.provider == "ollama"
+
+    def test_default_parsers_plugins(self):
+        cfg = rs.RuntimeSettings()
+        assert cfg.parsers.ocr.plugin == "rapidocr"
+        assert cfg.parsers.pdf.plugin == "pypdf"
+        assert cfg.parsers.docx.plugin == "python_docx"
+
+    def test_default_dedup(self):
+        cfg = rs.RuntimeSettings()
+        assert cfg.dedup.strategy == "sha256"
+        assert cfg.dedup.simhash_threshold == 3
+        assert cfg.dedup.ttl_seconds == 86400
+
+    def test_default_with_env_fallback_uses_ollama_env(self):
+        """无持久化文件时，base_url/model 取自 settings.ollama_*"""
+        cfg = rs._default_with_env_fallback()
+        from app.config import settings
+
+        assert cfg.models.summarize.base_url == settings.ollama_base_url
+        assert cfg.models.summarize.model == settings.ollama_model
+        assert cfg.models.query.base_url == settings.ollama_base_url
+        assert cfg.models.classify.model == settings.ollama_model
+
+
+# ---------------------------------------------------------------------------- #
+# 加载
+# ---------------------------------------------------------------------------- #
+
+
+class TestLoad:
+    def test_missing_file_returns_env_fallback(self, isolated_settings_path: Path):
+        """文件不存在时回退到带 env 兜底的默认值（不抛异常）"""
+        assert not isolated_settings_path.exists()
+        cfg = rs.load_runtime_settings()
+        # base_url 应来自 env fallback
+        from app.config import settings
+
+        assert cfg.models.summarize.base_url == settings.ollama_base_url
+
+    def test_valid_file_parsed(self, isolated_settings_path: Path):
+        isolated_settings_path.write_text(
+            json.dumps(
+                {
+                    "models": {
+                        "summarize": {"provider": "openai_compatible", "model": "gpt-4o-mini", "api_key": "k"}
+                    },
+                    "dedup": {"strategy": "simhash", "simhash_threshold": 5},
+                }
+            ),
+            encoding="utf-8",
+        )
+        cfg = rs.load_runtime_settings()
+        assert cfg.models.summarize.provider == "openai_compatible"
+        assert cfg.models.summarize.model == "gpt-4o-mini"
+        assert cfg.models.summarize.api_key == "k"
+        assert cfg.dedup.strategy == "simhash"
+        assert cfg.dedup.simhash_threshold == 5
+        # 未指定的字段保留默认
+        assert cfg.dedup.ttl_seconds == 86400
+        assert cfg.models.query.provider == "ollama"
+
+    def test_corrupted_file_falls_back(self, isolated_settings_path: Path):
+        isolated_settings_path.write_text("not-json{", encoding="utf-8")
+        cfg = rs.load_runtime_settings()
+        # 回退到默认（ollama provider）
+        assert cfg.models.summarize.provider == "ollama"
+
+    def test_invalid_values_falls_back(self, isolated_settings_path: Path):
+        """字段值非法（如未知 provider）时整体回退默认"""
+        isolated_settings_path.write_text(
+            json.dumps({"models": {"summarize": {"provider": "unknown_provider"}}}),
+            encoding="utf-8",
+        )
+        cfg = rs.load_runtime_settings()
+        assert cfg.models.summarize.provider == "ollama"  # 回退默认
+
+
+# ---------------------------------------------------------------------------- #
+# 保存
+# ---------------------------------------------------------------------------- #
+
+
+class TestSave:
+    def test_save_writes_valid_json(self, isolated_settings_path: Path):
+        cfg = rs.RuntimeSettings()
+        cfg.dedup.strategy = "simhash"
+        rs.save_runtime_settings(cfg)
+        assert isolated_settings_path.exists()
+        data = json.loads(isolated_settings_path.read_text(encoding="utf-8"))
+        assert data["dedup"]["strategy"] == "simhash"
+
+    def test_save_creates_parent_dir(self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
+        path = tmp_path / "nested" / "deep" / "runtime_settings.json"
+        monkeypatch.setenv("RUNTIME_SETTINGS_PATH", str(path))
+        rs.save_runtime_settings(rs.RuntimeSettings())
+        assert path.exists()
+
+    def test_save_atomic_no_tmp_left(self, isolated_settings_path: Path):
+        """保存后同目录无残留 .tmp 临时文件"""
+        rs.save_runtime_settings(rs.RuntimeSettings())
+        tmps = list(isolated_settings_path.parent.glob(".runtime_settings.*.tmp"))
+        assert tmps == []
+
+
+# ---------------------------------------------------------------------------- #
+# 单例 + reload
+# ---------------------------------------------------------------------------- #
+
+
+class TestSingleton:
+    def test_get_returns_singleton(self, isolated_settings_path: Path):
+        cfg1 = rs.get_runtime_settings()
+        cfg2 = rs.get_runtime_settings()
+        assert cfg1 is cfg2
+
+    def test_reload_rereads_disk(self, isolated_settings_path: Path):
+        """reload 强制重新读盘，单例替换为新对象"""
+        cfg1 = rs.get_runtime_settings()
+        # 直接改盘上文件
+        isolated_settings_path.write_text(
+            json.dumps({"dedup": {"strategy": "none"}}), encoding="utf-8"
+        )
+        cfg2 = rs.reload_runtime_settings()
+        assert cfg2 is not cfg1
+        assert cfg2.dedup.strategy == "none"
+
+
+# ---------------------------------------------------------------------------- #
+# 部分更新（深合并）
+# ---------------------------------------------------------------------------- #
+
+
+class TestUpdate:
+    def test_partial_update_models_summarize(self, isolated_settings_path: Path):
+        rs.get_runtime_settings()  # 初始化单例
+        new_cfg = rs.update_runtime_settings(
+            {"models": {"summarize": {"model": "qwen2.5:3b"}}}
+        )
+        assert new_cfg.models.summarize.model == "qwen2.5:3b"
+        # 其他字段保留
+        assert new_cfg.models.summarize.provider == "ollama"
+        assert new_cfg.models.query.provider == "ollama"
+
+    def test_partial_update_dedup(self, isolated_settings_path: Path):
+        rs.get_runtime_settings()
+        new_cfg = rs.update_runtime_settings(
+            {"dedup": {"strategy": "simhash", "simhash_threshold": 5}}
+        )
+        assert new_cfg.dedup.strategy == "simhash"
+        assert new_cfg.dedup.simhash_threshold == 5
+        # ttl 未在 patch 中，保留默认
+        assert new_cfg.dedup.ttl_seconds == 86400
+
+    def test_update_persists_to_disk(self, isolated_settings_path: Path):
+        rs.get_runtime_settings()
+        rs.update_runtime_settings({"dedup": {"strategy": "none"}})
+        data = json.loads(isolated_settings_path.read_text(encoding="utf-8"))
+        assert data["dedup"]["strategy"] == "none"
+
+    def test_update_replaces_singleton(self, isolated_settings_path: Path):
+        old = rs.get_runtime_settings()
+        new = rs.update_runtime_settings({"dedup": {"strategy": "none"}})
+        assert new is not old
+        # 后续 get 拿到的是新单例
+        assert rs.get_runtime_settings() is new
+
+    def test_update_empty_patch_keeps_all(self, isolated_settings_path: Path):
+        """空 patch 不改变任何字段"""
+        rs.get_runtime_settings()
+        new_cfg = rs.update_runtime_settings({})
+        assert new_cfg.dedup.strategy == "sha256"
+
+
+# ---------------------------------------------------------------------------- #
+# 重置
+# ---------------------------------------------------------------------------- #
+
+
+class TestReset:
+    def test_reset_returns_defaults(self, isolated_settings_path: Path):
+        # 先污染
+        rs.get_runtime_settings()
+        rs.update_runtime_settings({"dedup": {"strategy": "none"}})
+        assert rs.get_runtime_settings().dedup.strategy == "none"
+        # 重置
+        cfg = rs.reset_runtime_settings()
+        assert cfg.dedup.strategy == "sha256"
+        assert cfg.dedup.simhash_threshold == 3
+        assert cfg.parsers.ocr.plugin == "rapidocr"
+
+    def test_reset_persists_to_disk(self, isolated_settings_path: Path):
+        rs.get_runtime_settings()
+        rs.update_runtime_settings({"dedup": {"strategy": "none"}})
+        rs.reset_runtime_settings()
+        data = json.loads(isolated_settings_path.read_text(encoding="utf-8"))
+        assert data["dedup"]["strategy"] == "sha256"
+
+
+# ---------------------------------------------------------------------------- #
+# 深合并工具函数
+# ---------------------------------------------------------------------------- #
+
+
+class TestDeepMerge:
+    def test_nested_dict_merged(self):
+        target = {"a": {"b": 1, "c": 2}, "d": 3}
+        rs._deep_merge(target, {"a": {"b": 10}})
+        assert target == {"a": {"b": 10, "c": 2}, "d": 3}
+
+    def test_non_dict_overrides(self):
+        target = {"a": {"b": 1}}
+        rs._deep_merge(target, {"a": 99})
+        assert target == {"a": 99}
+
+    def test_new_key_added(self):
+        target = {"a": 1}
+        rs._deep_merge(target, {"b": 2})
+        assert target == {"a": 1, "b": 2}