Fix CSE crash on string tensor attributes

The CSE optimizer crashed with ORT_ENFORCE when encountering a node
with a string tensor attribute (e.g., LabelEncoder with keys_tensor
of type STRING). Replace the enforce with an early return of hash 0,
treating such nodes as distinct (consistent with AreScalarTensorAttributeEqual).

Fixes #28413

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: Rajat Monga

onnxruntime/core/optimizer/common_subexpression_elimination.cc

@@ -224,12 +224,14 @@ bool AreEqual(const ONNX_NAMESPACE::AttributeProto& lhs, const ONNX_NAMESPACE::A
   return false;
 }
 
-// Support scalar tensor attribute only for now.
+// Support scalar tensor attribute only for now. String tensors are not supported and return 0.
 std::size_t GetTensorAttributeHash(const ONNX_NAMESPACE::TensorProto& attr_t) {
   std::size_t hash = 0;
   if (utils::HasDataType(attr_t) && attr_t.dims_size() == 1 && attr_t.dims()[0] == 1) {
     int data_type = attr_t.data_type();
-    ORT_ENFORCE(data_type != onnx::TensorProto_DataType_STRING, "Unexpected tensor string type");
+    if (data_type == onnx::TensorProto_DataType_STRING) {
+      return hash;
+    }
     std::vector<uint8_t> unpacked_tensor;
     if (utils::UnpackInitializerData(attr_t, unpacked_tensor).IsOK()) {
       UpdateHash(data_type, hash);