feat: 完善视频传播、标注编辑和拆帧闭环

- 接入 SAM2 视频传播能力：新增 /api/ai/propagate，支持用当前帧 mask/polygon/bbox 作为 seed，通过 SAM2 video predictor 向前、向后或双向传播，并可保存为真实 annotation。
- 接入 SAM3 video tracker：通过独立 Python 3.12 external worker 调用 SAM3 video predictor/tracker，使用本地 checkpoint 与 bbox seed 执行视频级跟踪，并在模型状态中标记 video_track 能力。
- 完善 SAM 模型分发：sam_registry 按 model_id 明确区分 sam2 propagation 与 sam3 video_track，避免两个模型链路混用。
- 打通前端“传播片段”：VideoWorkspace 使用当前选中 mask 和当前 AI 模型调用后端传播接口，传播结果回写并刷新工作区已保存标注。
- 增强 SAM3 本地 checkpoint 配置：新增 sam3_checkpoint_path 配置和 .env.example 示例，状态检查改为基于本地 checkpoint/独立环境/模型包可用性。
- 完善视频拆帧参数：/api/media/parse 支持 parse_fps、max_frames、target_width，后端任务保存帧时间戳、源帧号和 frame_sequence 元数据。
- 增加运行时 schema 兼容处理：启动时为旧 frames 表补充 timestamp_ms 和 source_frame_number 列，避免旧库升级后缺字段。
- 强化 Canvas 标注编辑：补齐多边形闭合、点工具、顶点拖拽、边中点插入、Delete/Backspace 删除、区域合并和重叠去除等交互。
- 增强语义分类联动：选中 mask 后可通过右侧语义分类树更新标签、颜色和 class metadata，并同步到保存/导出链路。
- 增加关键帧时间轴体验：FrameTimeline 显示具体时间信息，并支持键盘左右方向键切换关键帧。
- 完善 AI 交互分割参数：前端保留正向点、反向点、框选和 interactive prompt 的调用状态，支持 SAM2 细化候选区域与 SAM3 bbox 入口。
- 扩展后端/前端 API 类型：新增 propagateMasks、传播请求/响应 schema，并补齐 annotation、导出、模型状态和任务接口的测试覆盖。
- 更新项目文档：同步 README、AGENTS、接口契约、需求冻结、设计冻结、前端元素审计、实施计划和测试计划，标明真实功能边界与剩余风险。
- 增加测试覆盖：补充 SAM2/SAM3 传播、SAM3 状态、媒体拆帧参数、Canvas 编辑、语义标签切换、时间轴、工作区传播和 API 合约测试。
- 加强仓库安全边界：将 sam3权重/ 加入 .gitignore，避免本地模型权重被误提交。

验证：npm run test:run；pytest backend/tests；npm run lint；npm run build；python -m py_compile；git diff --check。

backend/tests/test_media.py

@@ -84,6 +84,12 @@ def test_parse_media_queues_background_task(client, monkeypatch):
     assert data["progress"] == 0
     assert data["project_id"] == project["id"]
     assert data["celery_task_id"] == "celery-1"
+    assert data["payload"] == {
+        "source_type": "video",
+        "parse_fps": 5.0,
+        "max_frames": None,
+        "target_width": 640,
+    }
     assert queued == [data["id"]]
     assert published == [data["id"]]
 
@@ -94,6 +100,35 @@ def test_parse_media_queues_background_task(client, monkeypatch):
     assert project_detail["status"] == "parsing"
 
 
+def test_parse_media_accepts_frame_sequence_options(client, monkeypatch):
+    project = client.post("/api/projects", json={
+        "name": "Parse Options",
+        "video_path": "uploads/1/clip.mp4",
+        "source_type": "video",
+        "parse_fps": 30,
+    }).json()
+
+    class FakeAsyncResult:
+        id = "celery-options"
+
+    monkeypatch.setattr("routers.media.parse_project_media.delay", lambda task_id: FakeAsyncResult())
+    monkeypatch.setattr("routers.media.publish_task_progress_event", lambda task: None)
+
+    response = client.post(
+        f"/api/media/parse?project_id={project['id']}&parse_fps=15&max_frames=120&target_width=960"
+    )
+
+    assert response.status_code == 202
+    data = response.json()
+    assert data["payload"] == {
+        "source_type": "video",
+        "parse_fps": 15.0,
+        "max_frames": 120,
+        "target_width": 960,
+    }
+    assert client.get(f"/api/projects/{project['id']}").json()["parse_fps"] == 15.0
+
+
 def test_parse_task_runner_registers_frames(client, db_session, monkeypatch, tmp_path):
     from models import ProcessingTask
     from services.media_task_runner import run_parse_media_task
@@ -118,10 +153,14 @@ def test_parse_task_runner_registers_frames(client, db_session, monkeypatch, tmp
     frame_file.write_bytes(b"fake image")
 
     monkeypatch.setattr("services.media_task_runner.download_file", lambda object_name: b"video")
-    monkeypatch.setattr("services.media_task_runner.parse_video", lambda local_path, output_dir, fps: ([str(frame_file)], 25.0))
+    monkeypatch.setattr(
+        "services.media_task_runner.parse_video",
+        lambda local_path, output_dir, fps, max_frames=None, target_width=640: ([str(frame_file)], 25.0),
+    )
     monkeypatch.setattr("services.media_task_runner.extract_thumbnail", lambda local_path, thumbnail_path: open(thumbnail_path, "wb").write(b"thumb"))
     monkeypatch.setattr("services.media_task_runner.upload_file", lambda *args, **kwargs: None)
     monkeypatch.setattr("services.media_task_runner.upload_frames_to_minio", lambda frame_files, project_id: [f"projects/{project_id}/frames/frame_000001.jpg"])
+    monkeypatch.setattr("routers.projects.get_presigned_url", lambda object_name, expires=3600: f"http://storage/{object_name}")
     published = []
     monkeypatch.setattr(
         "services.media_task_runner.publish_task_progress_event",
@@ -131,6 +170,17 @@ def test_parse_task_runner_registers_frames(client, db_session, monkeypatch, tmp
     result = run_parse_media_task(db_session, task.id)
 
     assert result["frames_extracted"] == 1
+    assert result["frame_sequence"] == {
+        "original_fps": 25.0,
+        "parse_fps": 5.0,
+        "frame_count": 1,
+        "duration_ms": 0.0,
+        "target_width": 640,
+        "frame_width": None,
+        "frame_height": None,
+        "max_frames": None,
+        "object_prefix": f"projects/{project['id']}/frames",
+    }
     db_session.refresh(task)
     assert task.status == "success"
     assert task.progress == 100
@@ -140,6 +190,8 @@ def test_parse_task_runner_registers_frames(client, db_session, monkeypatch, tmp
     assert project_detail["status"] == "ready"
     frames = client.get(f"/api/projects/{project['id']}/frames").json()
     assert "frame_000001.jpg" in frames[0]["image_url"]
+    assert frames[0]["timestamp_ms"] == 0.0
+    assert frames[0]["source_frame_number"] == 0
 
 
 def test_parse_task_runner_skips_already_cancelled_task(db_session):