Fuse writing KV buffer into rope kernel (part 1: sgl-kernel) (#9077)

sgl-kernel/python/sgl_kernel/__init__.py

@@ -21,6 +21,7 @@ from sgl_kernel.attention import (
 )
 from sgl_kernel.cutlass_moe import cutlass_w4a8_moe_mm, get_cutlass_w4a8_moe_mm_data
 from sgl_kernel.elementwise import (
+    FusedSetKVBufferArg,
     apply_rope_with_cos_sin_cache_inplace,
     fused_add_rmsnorm,
     gelu_and_mul,