[Optimizer] Fix reinterpretation of strings in _get_numpy_value (#2514)

Signed-off-by: Christoph Berganski <christoph.berganski@gmail.com>

Author: iksnagreb

onnxscript/optimizer/_constant_folding.py

@@ -278,9 +278,18 @@ def _get_numpy_value(
         if size_limit is not None and const_value.size > size_limit:
             return None
         try:
-            # Reinterpret the array with `.view()` because some implementations of
-            # ir.TensorProtocol (e.g. PyTorch<=2.7) do not use ml_dtypes for bfloat16 etc.
-            array = const_value.numpy().view(const_value.dtype.numpy())
+            # Turn the constant value into a numpy array representation with the
+            # specifics of this conversion handled by the tensor type
+            array = const_value.numpy()
+            # Can/should not reinterpret strings via .view, resulting in
+            #   "TypeError: Cannot change data-type for array of references."
+            # There is also no reason to reinterpret strings, this is only
+            # relevant for some arithmetic types
+            if const_value.dtype != ir.DataType.STRING:
+                # Reinterpret the array with `.view()` because some
+                # implementations  of ir.TensorProtocol (e.g. PyTorch<=2.7) do
+                # not use ml_dtypes for bfloat16 etc.
+                array = array.view(const_value.dtype.numpy())
         except FileNotFoundError:
             # External data is not available.
             logger.warning(