[Core] Performance: Use list[np.ndarray] instead of list[list[int]] for output tokens for GC optimization (#26368)

Signed-off-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>

tests/v1/spec_decode/test_eagle.py

@@ -3,6 +3,7 @@
 
 from unittest import mock
 
+import numpy as np
 import pytest
 import torch
 
@@ -112,7 +113,9 @@ def test_prepare_next_token_ids():
     sampled_token_ids_tensor = torch.tensor(
         sampled_token_ids, dtype=torch.int32, device=device
     )
-    sampled_token_ids_cpu = [[i for i in seq if i != -1] for seq in sampled_token_ids]
+    sampled_token_ids_cpu = [
+        np.array([i for i in seq if i != -1]) for seq in sampled_token_ids
+    ]
 
     expected_next_token_ids_cpu = [1, 4, 30, 40]
     expected_next_token_ids_tensor = torch.tensor(