[torchlib] Fix aten_instance_norm (#1964)

Otherwise exporter raises `Could not determine the dtype for the input
'inputs'.`

onnxscript/function_libs/torch_lib/ops/core.py

@@ -4391,7 +4391,10 @@ def aten_instance_norm(
     ), "running_mean and running_var must be provided when use_input_stats is False"
 
     batch_size = op.Shape(input, start=0, end=1)
-    bn_input = op.Reshape(input, op.Concat([1, -1], op.Shape(input, start=2), axis=0))
+    bn_input = op.Reshape(
+        input,
+        op.Concat(op.Constant(value_ints=[1, -1]), op.Shape(input, start=2), axis=0),
+    )
     weight = op.Tile(weight, batch_size)
     bias = op.Tile(bias, batch_size)
     running_mean = op.Tile(running_mean, batch_size)