Add initial support for RotaryEmbedding fusion for onnx opset 23 (#2450)

Add initial support for RotaryEmbedding fusion for onnx opset 23

---------

Signed-off-by: Ganesan Ramalingam <[email protected]>

Author: gramalingam

…iter/ort_fusions/models/_bart_encoder.py …xscript/rewriter/models/_bart_encoder.pyonnxscript/rewriter/ort_fusions/models/_bart_encoder.py renamed to onnxscript/rewriter/models/_bart_encoder.py onnxscript/rewriter/ort_fusions/models/_bart_encoder.py renamed to onnxscript/rewriter/models/_bart_encoder.py

@@ -4,7 +4,7 @@
 
 import onnx_ir as ir
 
-from onnxscript.rewriter.onnx_fusions import _rms_normalization
+from onnxscript.rewriter.onnx_fusions import _rms_normalization, _rotary_embedding
 
 
 def _get_onnx_opset_version(model: ir.Model) -> int | None:
@@ -23,6 +23,7 @@ def _opset_23_fuse(model: ir.Model, *, debug: bool = False) -> dict[str, int]:
     """Apply fusions targeting ONNX opset 23."""
     counts: dict[str, int] = {}
     counts["RMSNormalization"] = _rms_normalization.fuse_rms_normalization(model, debug=debug)
+    counts["RotaryEmbedding"] = _rotary_embedding.fuse_rotary_embedding(model, debug=debug)
     return counts