TFX support for TFDS

Is your feature request related to a problem? Please describe.
TFX is an end-to-end platform for deploying production ML pipelines.
They have an ExampleGen to load data from various sources (databases, tfrecord,...):
It would be good to have ExampleGen support TFDS natively.
This would unlock TFDS datasets for production models.

Describe the solution you'd like
A possible solution could be to reuse the existing TFRecord writer in TFDS and create an API
like:

example_gen = CreateExampleGenfromTFDS(name="cifar10")

which would somewhat mimic the following code:

Sample Code using TFX with TFDS

import tensorflow_datasets as tfds
from tfx.utils.dsl_utils import external_input
from tfx.components.example_gen.import_example_gen.component import ImportExampleGen

# Generate TFRecords with TFDS
builder = tfds.builder("cifar10")
builder.download_and_prepare()

# Use the TFRecords to create a ExampleGen
examples = external_input(builder.data_dir)
example_gen = ImportExampleGen(input=examples)

Describe alternatives you've considered

• Reuse ExempleGen or develop our own

Start with the TFX ExampleGen documentation and the TFDS TFRecord writer referenced in the issue, then compare the sample builder.download_and_prepare and ImportExampleGen flow. Define the integration boundary and API needed for native TFDS input; done means TFDS datasets can be used by production TFX pipelines without the manual external_input and ImportExampleGen workaround.