Add read_avro and list_avro_columns for rework on Splittable Avro sup…

…port (#399)

This PR is part of the effort to rework on Dataset with large files reading into Tensors first to speed up performance. See 382 and 366 for related discussions.

Summary:

1) read_avro is able to read a avro file within the range of [offset, offset+length] (Splittable)
2) we use primitive read_avro C++ ops to read in big chunks and then wire up with tf.data.Dataset
3) read_avro could be used in other places.
4) AvroDataset automatically find out the dtype in eager mode, in graph mode, user has
   to specify the dtype in kwargs.

Signed-off-by: Yong Tang <yong.tang.github@outlook.com>

tensorflow_io/avro/BUILD

@@ -10,7 +10,7 @@ load(
 cc_library(
     name = "avro_ops",
     srcs = [
-        "kernels/avro_input.cc",
+        "kernels/avro_kernels.cc",
         "ops/avro_ops.cc",
     ],
     copts = tf_io_copts(),

tensorflow_io/avro/__init__.py

@@ -15,18 +15,24 @@
 """Avro Dataset.
 
 @@AvroDataset
+@@list_avro_columns
+@@read_avro
 """
 
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
 from tensorflow_io.avro.python.ops.avro_ops import AvroDataset
+from tensorflow_io.avro.python.ops.avro_ops import list_avro_columns
+from tensorflow_io.avro.python.ops.avro_ops import read_avro
 
 from tensorflow.python.util.all_util import remove_undocumented
 
 _allowed_symbols = [
     "AvroDataset",
+    "list_avro_columns",
+    "read_avro_",
 ]
 
 remove_undocumented(__name__, allowed_exception_list=_allowed_symbols)