Fix rlds episode_metadata

[Tavish9]

This PR fixes rlds's build_info when episode_metadata is not None

Reproduce

from tensorflow_datasets.rlds import rlds_base
features = rlds_base.build_info(
    rlds_base.DatasetConfig(
        name="xxx",
        steps={xxx},
        episode_metadata_info={
            "test": tfds.features.Text()
        }
    ),
    self,
)
print(features)

FeaturesDict({
    'steps': Dataset({
        ...
    }),
    'test': Text(shape=(), dtype=string),
}),

Currently, the returned features will get rid of episode_metadata key, which is not consistent with widely used rlds dataset:

• robomimic
• bridge
• ...
• All datasets listed in official tensorflow datasets robotics

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[fineguy]

@Tavish9 thank for the PR!

IIUC, this changes the dataset info, so you need to change the example generation as well, see the failed tests.

[Tavish9]

@fineguy already fix

[fineguy]

@Tavish9 please fix the failed tests.

[Tavish9]

I also updated the logic of generate_examples, please check

[fineguy]

@Tavish9 please confirm locally that tests run successfully, this will be easier for the review.

[Tavish9]

@fineguy sorry for bothering you. I don't find any steps for local testing in README.

Could you please show me the way?

[fineguy]

@Tavish9 check this documentation: https://www.tensorflow.org/datasets/contribute#add_and_run_unittests

[Tavish9]

Already passed all tests in rlds.

[Tavish9]

@fineguy take a look, please