Specify TokenTextEncoder with custom tokenizer makes dataset reusing failed

Short description
Description of the bug.

Environment information

• Operating System: Ubuntu 16.04
• Python version: 3.6.8
• tensorflow-datasets/tfds-nightly version: 1.0.2
• tensorflow/tensorflow-gpu/tf-nightly/tf-nightly-gpu version: 1.13.1

Reproduction instructions

 BUILDER_CONFIGS = [
      DatasetConfig(
          name="words",
          version="1.0.0",
          description=("Uses word-level text encoding with "
                       "`tfds.features.text.TokenTextEncoder`"),
          text_encoder_config=tfds.features.text.TextEncoderConfig(
              encoder=tfds.features.text.TokenTextEncoder(
                  vocab_list=VOCAB,
                  tokenizer=Tokenizer(reserved_tokens=VOCAB))),
      ),
  ]

Link to logs

tensorflow.python.framework.errors_impl.NotFoundError: tensorflow_datasets/tacotron_dataset/phones/1.0.0.incompleteGTGY8F/text.text.tokens.tokenizer; No such file or directory

Expected behavior
Dataset can be reused when rerunning.

Additional context
When custom tokenizer is specified, it's saved to file for reusing. (https://github.com/tensorflow/datasets/blob/master/tensorflow_datasets/core/features/text/text_encoder.py#L331). But when reusing happens, it seems that wrong path is used to load tokenizer from file.

if self._user_defined_tokenizer is not None:
      self._tokenizer.save_to_file(filename)
      kwargs["tokenizer_file_prefix"] = filename

Inspect tensorflow_datasets/core/features/text/text_encoder.py around the custom tokenizer save and load logic, then reproduce the failure with the TokenTextEncoder and custom Tokenizer configuration shown. Done means a dataset using this configuration can be reused on rerun without the tokenizer path NotFoundError.