完善项目导入、模板与分割工作区交互

- 增强 DICOM/视频项目导入与演示数据：DICOM 按文件名自然顺序处理，导入后展示上传与解析任务进度，恢复演示出厂设置保留演示视频和演示 DICOM 项目，并补充 demo media seed 逻辑。

- 完善项目管理：项目支持重命名、删除、复制，删除使用站内确认弹窗，复制支持新项目重置和全内容复制，DICOM 项目不显示生成帧入口。

- 完善 GT Mask 与导出链路：只支持 8-bit maskid 图导入，非法/全背景图明确拒绝，尺寸自动适配，高精度 polygon 回显；统一导出默认当前帧，GT_label 使用 uint8 和真实 maskid，待分类 maskid 0 与背景一致。

- 完善分割工作区交互：新增画笔和橡皮擦并支持尺寸控制，移除创建点/线段入口，工具栏按类别分隔，AI 智能分割使用明确 AI 图标，取消黄色 seed point，清空/删除传播 mask 后同步清理空帧时间轴状态。

- 完善传播与时间轴：自动传播使用 SAM 2.1 权重任务，参考帧无遮罩时提示，传播历史按同一蓝色系递进变暗，删除/清空传播链时保留人工或独立 AI 标注来源。

- 完善模板库：新增头颈部 CT 分割默认模板，所有模板保留 maskid 0 待分类，支持鼠标复制模板、拖拽层级、JSON 批量导入预览、删除 label 和站内删除确认。

- 完善用户与高风险确认：用户改密码、删除用户、恢复演示出厂设置和清空人工/AI 标注帧均改为站内确认交互，避免浏览器原生 prompt/confirm。

- 补充前后端测试与文档：更新项目、模板、GT 导入、导出、传播、DICOM、用户管理等测试，并同步 README、AGENTS 和 doc 下实现/契约/测试计划文档。

backend/tests/test_media.py

@@ -48,15 +48,37 @@ def test_upload_dicom_batch_filters_files_and_creates_project(client, monkeypatc
     response = client.post(
         "/api/media/upload/dicom",
         files=[
-            ("files", ("a.dcm", b"dcm", "application/dicom")),
+            ("files", ("10.dcm", b"dcm10", "application/dicom")),
             ("files", ("skip.txt", b"text", "text/plain")),
+            ("files", ("2.dcm", b"dcm2", "application/dicom")),
+            ("files", ("1.dcm", b"dcm1", "application/dicom")),
         ],
     )
 
     assert response.status_code == 201
     data = response.json()
-    assert data["uploaded_count"] == 1
-    assert uploaded == [f"uploads/{data['project_id']}/dicom/a.dcm"]
+    assert data["uploaded_count"] == 3
+    assert uploaded == [
+        f"uploads/{data['project_id']}/dicom/1.dcm",
+        f"uploads/{data['project_id']}/dicom/2.dcm",
+        f"uploads/{data['project_id']}/dicom/10.dcm",
+    ]
+    project_detail = client.get(f"/api/projects/{data['project_id']}").json()
+    assert project_detail["name"] == "1.dcm"
+
+
+def test_upload_dicom_batch_rejects_when_no_valid_dicom(client, monkeypatch):
+    monkeypatch.setattr("routers.media.upload_file", lambda *args, **kwargs: None)
+
+    response = client.post(
+        "/api/media/upload/dicom",
+        files=[
+            ("files", ("notes.txt", b"text", "text/plain")),
+        ],
+    )
+
+    assert response.status_code == 400
+    assert response.json()["detail"] == "No valid DICOM files uploaded"
 
 
 def test_parse_media_queues_background_task(client, monkeypatch):
@@ -194,6 +216,101 @@ def test_parse_task_runner_registers_frames(client, db_session, monkeypatch, tmp
     assert frames[0]["source_frame_number"] == 0
 
 
+def test_parse_dicom_reads_files_in_natural_filename_order(monkeypatch, tmp_path):
+    from pathlib import Path
+
+    import numpy as np
+
+    from services.frame_parser import parse_dicom
+
+    dcm_dir = tmp_path / "dcm"
+    output_dir = tmp_path / "frames"
+    dcm_dir.mkdir()
+    for name in ["10.dcm", "2.dcm", "1.dcm"]:
+        (dcm_dir / name).write_bytes(b"dcm")
+
+    read_order = []
+
+    class FakeDicom:
+        pixel_array = np.ones((2, 2), dtype=np.uint8)
+
+    def fake_dcmread(path):
+        read_order.append(Path(path).name)
+        return FakeDicom()
+
+    def fake_imwrite(path, image, params=None):
+        Path(path).write_bytes(image.tobytes())
+        return True
+
+    monkeypatch.setattr("services.frame_parser.dcmread", fake_dcmread)
+    monkeypatch.setattr("services.frame_parser.cv2.imwrite", fake_imwrite)
+
+    frame_files = parse_dicom(str(dcm_dir), str(output_dir))
+
+    assert read_order == ["1.dcm", "2.dcm", "10.dcm"]
+    assert [Path(path).name for path in frame_files] == ["frame_000000.jpg", "frame_000001.jpg", "frame_000002.jpg"]
+
+
+def test_parse_task_runner_downloads_dicom_objects_in_natural_filename_order(client, db_session, monkeypatch, tmp_path):
+    from types import SimpleNamespace
+
+    from models import ProcessingTask
+    from services.media_task_runner import run_parse_media_task
+
+    project = client.post("/api/projects", json={
+        "name": "DICOM",
+        "video_path": "uploads/1/dicom",
+        "source_type": "dicom",
+        "parse_fps": 30,
+    }).json()
+    task = ProcessingTask(
+        task_type="parse_dicom",
+        status="queued",
+        progress=0,
+        project_id=project["id"],
+        payload={"source_type": "dicom"},
+    )
+    db_session.add(task)
+    db_session.commit()
+    db_session.refresh(task)
+
+    class FakeClient:
+        def list_objects(self, bucket, prefix, recursive=True):
+            return [
+                SimpleNamespace(object_name=f"{prefix}/10.dcm"),
+                SimpleNamespace(object_name=f"{prefix}/2.dcm"),
+                SimpleNamespace(object_name=f"{prefix}/1.dcm"),
+            ]
+
+    downloaded = []
+    frame_files = []
+    for idx in range(3):
+        frame_file = tmp_path / f"frame_{idx:06d}.jpg"
+        frame_file.write_bytes(b"fake image")
+        frame_files.append(str(frame_file))
+
+    monkeypatch.setattr("services.media_task_runner.get_minio_client", lambda: FakeClient())
+    monkeypatch.setattr(
+        "services.media_task_runner.download_file",
+        lambda object_name: downloaded.append(object_name) or b"dcm",
+    )
+    monkeypatch.setattr("services.media_task_runner.parse_dicom", lambda *args, **kwargs: frame_files)
+    monkeypatch.setattr(
+        "services.media_task_runner.upload_frames_to_minio",
+        lambda frames, project_id: [f"projects/{project_id}/frames/{idx}.jpg" for idx, _ in enumerate(frames)],
+    )
+    monkeypatch.setattr("services.media_task_runner.publish_task_progress_event", lambda task: None)
+
+    result = run_parse_media_task(db_session, task.id)
+
+    assert result["frames_extracted"] == 3
+    assert downloaded == [
+        "uploads/1/dicom/1.dcm",
+        "uploads/1/dicom/2.dcm",
+        "uploads/1/dicom/10.dcm",
+    ]
+
+
 def test_parse_task_runner_skips_already_cancelled_task(db_session):
     from models import ProcessingTask
     from services.media_task_runner import run_parse_media_task