[MUSA][18/N] Add MUSA-optimized kernel implementations for hot ops (#23255)

Signed-off-by: Joey-gvwal <joey_gvwal@yeah.net>
Co-authored-by: R0CKSTAR <yeahdongcn@gmail.com>

sgl-kernel/python/sgl_kernel/__init__.py

@@ -111,6 +111,15 @@ from sgl_kernel.version import __version__
 if torch.version.hip is not None:
     from sgl_kernel.elementwise import gelu_quick
 
+if hasattr(torch.version, "musa") and torch.version.musa is not None:
+    from sgl_kernel.musa import (
+        musa_batched_rotary_embedding_contiguous,
+        musa_fused_gemv,
+        musa_fused_moe_gemv,
+        musa_fused_mul_add,
+        musa_rotary_embedding_contiguous,
+    )
+
 
 _DEBUG_EXPORT_NAMES = [
     "apply_shuffle_mul_sum",