tiny update: use rope kernel in sgl-kernel for amd (#14955)

python/sglang/srt/layers/rotary_embedding.py

@@ -118,7 +118,7 @@ class RotaryEmbedding(CustomOp):
             and not (_is_cpu and _is_cpu_amx_available)
             and not (_is_xpu)
         ):
-            if _is_cuda:
+            if _is_cuda or _is_hip:
                 from sgl_kernel import rotary_embedding
             else:
                 from vllm._custom_ops import rotary_embedding