feat(simclr): 接入 SwanLab 可视化，并补上本地看板缺失的依赖

- SimCLRTrainer 记录 simclr/loss、loss_avg、lr（按 step）与 epoch_loss、
  epoch_seconds、gpu_mem_peak_mb（按 epoch），超参为完整 yaml 配置；
  正常结束与 KeyboardInterrupt 两条路径都会 finish()
- 配置新增 simclr.swanlab（字段同 train.swanlab），默认 mode: local，
  不需要登录，日志写到 <work_dir>/swanlog
- 依赖改为 swanlab[dashboard]：此前 `swanlab watch` 在干净环境里会报
  "Please install the swanboard package"，README 里检测训练那节的看板命令
  同样受影响，现已随 uv sync 装好
- README 补齐本地看板/云端/离线补传三种用法与端口、局域网参数

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

tests/test_simclr.py

@@ -104,6 +104,61 @@ def test_simclr_trainer_runs_and_saves_backbone_only(tiny_image_dir, tmp_path):
     backbone.load_state_dict(state["model"])  # 能被干净地加载回去
 
 
+def test_simclr_logs_metrics_to_swanlab(tiny_image_dir, tmp_path, monkeypatch):
+    """SwanLab 接线：按 step 记 loss/lr，按 epoch 记平均 loss、耗时与显存峰值，结束时收尾。"""
+    from torch.utils.data import DataLoader
+
+    import engine.simclr as simclr_mod
+    from datasets.unlabeled import UnlabeledImageDataset, unlabeled_collate
+
+    records: list[tuple[dict, int | None, str]] = []
+    finished: list[bool] = []
+
+    class StubLogger:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+        def log(self, metrics, step=None, prefix=""):
+            records.append((metrics, step, prefix))
+
+        def finish(self):
+            finished.append(True)
+
+    monkeypatch.setattr(simclr_mod, "SwanLabLogger", StubLogger)
+
+    dataset = UnlabeledImageDataset(
+        tiny_image_dir,
+        transforms=[{"name": "Resize", "size": [32, 32]}, {"name": "ToTensor"}],
+        views=2,
+    )
+    loader = DataLoader(dataset, batch_size=4, shuffle=True, drop_last=True, collate_fn=unlabeled_collate)
+    backbone = BACKBONES.build({"name": "tiny_conv", "widths": [8, 16], "depths": [1, 1]})
+
+    trainer = simclr_mod.SimCLRTrainer(
+        backbone,
+        loader,
+        {"epochs": 1, "amp": False, "log_interval": 1, "save_interval": 0,
+         "projection": {"hidden_channels": 16, "out_channels": 8},
+         "swanlab": {"enabled": True, "mode": "disabled"}},
+        torch.device("cpu"),
+        tmp_path / "run",
+        full_config={"method": "simclr"},
+    )
+    trainer.fit()
+
+    assert records, "没有任何指标被记录"
+    assert all(prefix == "simclr/" for _, _, prefix in records)
+    step_keys = {k for metrics, _, _ in records for k in metrics}
+    assert {"loss", "loss_avg", "lr"} <= step_keys
+    assert {"epoch", "epoch_loss", "epoch_seconds", "gpu_mem_peak_mb"} <= step_keys
+    assert all(step is not None and step > 0 for _, step, _ in records)
+    assert finished == [True]  # fit() 结束时调用了 finish()
+
+    # 完整 yaml 配置作为超参传给了 SwanLab
+    assert trainer.visualizer.kwargs["config"] == {"method": "simclr"}
+    assert trainer.visualizer.kwargs["mode"] == "disabled"
+
+
 def test_simclr_saves_intermediate_checkpoints(tiny_image_dir, tmp_path):
     """中途落盘：跑到一半被打断也要能拿到权重（此前只在全部 epoch 跑完后才存）。"""
     from torch.utils.data import DataLoader