pt: avoid D2H in se_e2_a (#3424)

sec is used as a slice index, so it should not stored on the GPU,
otherwise, D2H will happen to create the tensor with the shape.

Before:

![image](https://github.com/deepmodeling/deepmd-kit/assets/9496702/f5a520d8-ed83-4520-aed0-d8fed547c293)

After:

![image](https://github.com/deepmodeling/deepmd-kit/assets/9496702/5548632b-3099-4fe2-ab53-5c570abd714a)

Signed-off-by: Jinzhe Zeng <jinzhe.zeng@rutgers.edu>

deepmd/pt/model/descriptor/se_a.py

@@ -342,9 +342,8 @@ def __init__(
         self.reinit_exclude(exclude_types)
 
         self.sel = sel
-        self.sec = torch.tensor(
-            np.append([0], np.cumsum(self.sel)), dtype=int, device=env.DEVICE
-        )
+        # should be on CPU to avoid D2H, as it is used as slice index
+        self.sec = [0, *np.cumsum(self.sel).tolist()]
         self.split_sel = self.sel
         self.nnei = sum(sel)
         self.ndescrpt = self.nnei * 4