[Kernel] RFC #29630 finale: retire sglang.jit_kernel into sglang.kernels (#32072)

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

sgl-kernel/python/sgl_kernel/moe.py

@@ -106,7 +106,7 @@ def moe_sum(
 
 # moe_fused_gate / kimi_k2_moe_fused_gate (AOT gate kernels) retired — the gate/topk
 # path is consolidated onto the unified Triton router in
-# python/sglang/jit_kernel/moe_fused_gate.py (sglang issue #26771).
+# python/sglang/kernels/ops/moe/moe_fused_gate.py (sglang issue #26771).
 
 
 def fp8_blockwise_scaled_grouped_mm(

sgl-kernel/python/sgl_kernel/testing/rotary_embedding.py

@@ -3,8 +3,10 @@ from typing import Optional, Tuple, Union
 
 import torch
 
-from sglang.jit_kernel.rope import FusedSetKVBufferArg as _JitFusedSetKVBufferArg
-from sglang.jit_kernel.rope import (
+from sglang.kernels.ops.attention.rope import (
+    FusedSetKVBufferArg as _JitFusedSetKVBufferArg,
+)
+from sglang.kernels.ops.attention.rope import (
     apply_rope_with_cos_sin_cache_inplace as _jit_apply_rope_with_cos_sin_cache_inplace,
 )