[tfds] Remove config name text datasets

As TFDS text encoding API has many bug, performance issues and is unsupported, we are cleaning up our datasets to remove all encoder builder config and only keep the plain text version of the datasets.

To use encoder, please use `tensorflow_text` which is a more flexible, more supported, performant text encoding API.

For forward compatibility, `tfds.load('ds/plain_text')` -> `tfds.load('ds')` will load the default plain text dataset.

PiperOrigin-RevId: 326198682

Author: Conchylicultor

mesh_tensorflow/transformer/gin/problems/lm1b_untok.gin

@@ -5,7 +5,7 @@
 import mesh_tensorflow.transformer.t2t_vocabulary
 
 # Dataset
-dataset_name = "lm1b/plain_text"
+dataset_name = "lm1b"
 utils.run.train_dataset_fn = @mesh_tensorflow.transformer.dataset.untokenized_tfds_dataset
 dataset.untokenized_tfds_dataset.dataset_name = %dataset_name
 dataset.untokenized_tfds_dataset.text2self = True