Improve usability of batch_size in as_dataset function.

[ChanchalKumarMaji]

Currently, in the as_dataset function, batch_size=1 won't add a batch dimension while batch_size=2 will. This is inconsistent for the users as passing the batch_size as flag will modify the tensors shape.
Sol: Have batch_size=None (default) do not return batch_dimension while batch_size=1 does.

[ChanchalKumarMaji]

@Conchylicultor please check.

[ChanchalKumarMaji]

kokoro is giving an error.

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/tmpfs/src/github/datasets/tensorflow_datasets/__init__.py", line 46, in <module>
    from tensorflow_datasets.core import tf_compat
  File "/tmpfs/src/github/datasets/tensorflow_datasets/core/__init__.py", line 18, in <module>
    from tensorflow_datasets.core.dataset_builder import BeamBasedBuilder
  File "/tmpfs/src/github/datasets/tensorflow_datasets/core/dataset_builder.py", line 30, in <module>
    import tensorflow as tf
  File "/home/kbuilder/virtualenv/tfds_env/lib/python3.6/site-packages/tensorflow/__init__.py", line 93, in <module>
    from tensorflow_core import *
AttributeError: module 'tensorflow_core' has no attribute 'compiler'

Should I bother about it ?