Infer tensor representations from Features even if some TRs are already present in

TFXIOs.
The reason for this change is to avoid ignoring provided features when unrelated TRs are provided as well for a less confusing behaviour.

PiperOrigin-RevId: 487586847

Author: zoyahav

tensorflow_transform/beam/impl.py

@@ -925,21 +925,10 @@ class _InstanceDictInputToTFXIOInput(beam.PTransform):
 
   def __init__(self, schema, desired_batch_size):
     self._schema = schema
-    # Infer all features in the schema as TensorRepresentations. This is needed
-    # because TFXIO will ignore Features if TensorRepresentations are present in
-    # the schema.
-    # TODO(b/202791319) Investigate whether this should be done at TFXIO level.
-    tensor_representations = (
-        tensor_representation_util.InferTensorRepresentationsFromMixedSchema(
-            schema))
-    extended_schema = schema_pb2.Schema()
-    extended_schema.CopyFrom(schema)
-    tensor_representation_util.SetTensorRepresentationsInSchema(
-        extended_schema, tensor_representations)
     self._tfxio = tf_example_record.TFExampleBeamRecord(
         physical_format='inmem',
         telemetry_descriptors=['StandaloneTFTransform'],
-        schema=extended_schema)
+        schema=schema)
     self._desired_batch_size = desired_batch_size
 
   def tensor_adapter_config(self):