[XPU] Enable qwen3.5 on XPU (#21668)

Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>

test/registered/attention/test_chunk_gated_delta_rule.py

@@ -6,12 +6,17 @@ from sglang.srt.layers.attention.fla.chunk import chunk_gated_delta_rule
 from sglang.srt.layers.attention.fla.fused_recurrent import (
     fused_recurrent_gated_delta_rule,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.srt.utils import get_device
+from sglang.test.ci.ci_register import register_cuda_ci, register_xpu_ci
 
 register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
+register_xpu_ci(est_time=30, suite="xpu")
 
 
-@unittest.skipIf(not torch.cuda.is_available(), "Test requires CUDA")
+@unittest.skipIf(
+    not (torch.cuda.is_available() or torch.xpu.is_available()),
+    "Test requires CUDA or XPU",
+)
 class TestChunkGatedDeltaRule(unittest.TestCase):
     """Test chunk_gated_delta_rule against token-by-token fused_recurrent reference."""
 
@@ -68,7 +73,7 @@ class TestChunkGatedDeltaRule(unittest.TestCase):
         self, B, T_per_seq, H, K, V, pool_size, sequential_indices=False, seed=42
     ):
         """Run correctness check for one (B, T_per_seq, H, K, V, pool_size) config."""
-        device = "cuda"
+        device = get_device()
         dtype = torch.bfloat16
         T = B * T_per_seq