Unconsumed dataset iterator raises: AttributeError: 'NoneType' object has no attribute 'eager_mode'

Crash on exit.

• OS: Ubuntu 22.04

• Python version: 3.10

• tensorflow-datasets 4.5.2

• tensorflow 2.8.0

• Does the issue still exists with the last tfds-nightly package (pip install --upgrade tfds-nightly) ? Yes.

Reproduction instructions

import tensorflow_datasets as tfds
ds_map, info = tfds.load('iris', as_supervised=True, with_info=True, batch_size=1)
ds = ds_map['train']
dataset_iterator = iter(ds)

Produces

AttributeError: 'NoneType' object has no attribute 'eager_mode'

Start by running the provided tfds.load('iris', ...) reproduction and creating an iterator without consuming it. Trace the iterator cleanup path from that entry point to identify why eager_mode is accessed on None; done means exiting without the AttributeError and preserving the reported dataset behavior.