TFDSv4: store contextual information from datasets, such as sample rate and frame rate

[davidslater]

Currently, this information is needed in the following scenarios:

audio_asr.py:            self.test_dataset.context.sample_rate,
audio_classification.py:            self.test_dataset.context.sample_rate,
carla_video_tracking.py:            frame_rate=self.test_dataset.context.frame_rate,
carla_video_tracking.py:            frame_rate=self.test_dataset.context.frame_rate,
video_ucf101_scenario.py:            frame_rate=self.test_dataset.context.frame_rate,

This needs to be captured somewhere in the data loading pipeline.