[torchlib] Modify aten_unbind to use None for split_sizes (#2536)

According to
https://onnx.ai/onnx/operators/onnx__SplitToSequence.html#summary, `If
the argument split is not specified, a default scalar value of 1 is used
as the value of split`, and this is the only case when `keepdims` can be
set to `0`.

Fixes #2533

Author: justinchuby

onnxscript/function_libs/torch_lib/ops/core.py

@@ -8718,12 +8718,11 @@ def aten_type_as(self: TTensor, other: TTensor2) -> TTensor2:
     return op.CastLike(self, other)
 
 
-@torch_op("aten::unbind.int")
+@torch_op("aten::unbind.int", trace_only=True)
 def aten_unbind(self: TTensor, dim: int = 0) -> Sequence[TTensor]:
     """unbind.int(Tensor(a -> *) self, int dim=0) -> Tensor(a)[]"""
 
-    split_sizes = op.Constant(value_int=1)
-    return op.SplitToSequence(self, split_sizes, axis=dim, keepdims=False)
+    return op.SplitToSequence(self, axis=dim, keepdims=False)
 
 
 @torch_op("aten::unflatten.int", trace_only=True)