BERT Language Models: Training model metas no longer supported with TF2.x

[jthestness]

When migrating to TF2.x support, we're no longer able to load BERT training model metas. Caused by ops that have been removed from TF:

$ python catamount/tests/full/tf_bert_language_models.py --model_name uncased_L-3_H-768_A-12.training
WARN: Cannot import tensorflow.contrib.mpi_collectives... not built?
WARN: Cannot import tensorswift... not installed?
Traceback (most recent call last):
  File "catamount/tests/full/tf_bert_language_models.py", line 752, in <module>
    run_tf_bert_lm(model_name=args.model_name)
  File "catamount/tests/full/tf_bert_language_models.py", line 42, in run_tf_bert_lm
    graph = catamount.frameworks.tensorflow.import_graph(graph_meta)
  File "build/bdist.linux-x86_64/egg/catamount/frameworks/tensorflow.py", line 287, in import_graph
  File "build/bdist.linux-x86_64/egg/catamount/frameworks/tensorflow.py", line 276, in load_tf_session
  File "/net/joel-dev/srv/nfs/joel-data/ws/local/envs/py3.7.4_tf2.2.2/lib/python3.7/site-packages/tensorflow/python/training/saver.py", line 1462, in import_meta_graph
    **kwargs)[0]
  File "/net/joel-dev/srv/nfs/joel-data/ws/local/envs/py3.7.4_tf2.2.2/lib/python3.7/site-packages/tensorflow/python/training/saver.py", line 1486, in _import_meta_graph_with_return_elements
    **kwargs))
  File "/net/joel-dev/srv/nfs/joel-data/ws/local/envs/py3.7.4_tf2.2.2/lib/python3.7/site-packages/tensorflow/python/framework/meta_graph.py", line 799, in import_scoped_meta_graph_with_return_elements
    return_elements=return_elements)
  File "/net/joel-dev/srv/nfs/joel-data/ws/local/envs/py3.7.4_tf2.2.2/lib/python3.7/site-packages/tensorflow/python/util/deprecation.py", line 507, in new_func
    return func(*args, **kwargs)
  File "/net/joel-dev/srv/nfs/joel-data/ws/local/envs/py3.7.4_tf2.2.2/lib/python3.7/site-packages/tensorflow/python/framework/importer.py", line 405, in import_graph_def
    producer_op_list=producer_op_list)
  File "/net/joel-dev/srv/nfs/joel-data/ws/local/envs/py3.7.4_tf2.2.2/lib/python3.7/site-packages/tensorflow/python/framework/importer.py", line 497, in _import_graph_def_internal
    graph._c_graph, serialized, options)  # pylint: disable=protected-access
tensorflow.python.framework.errors_impl.NotFoundError: Op type not registered 'MapAndBatchDatasetV2' in binary running on joel-dev. Make sure the Op and Kernel are registered in the binary running in this process. Note that if you are loading a saved graph which used ops from tf.contrib, accessing (e.g.) `tf.contrib.resampler` should be done before importing the graph, as contrib ops are lazily registered when the module is first accessed.

Decide whether to update, deprecate, or otherwise fix.