[data] fix repartitioning empty datasets

[raulchen]

Fix the following error when repartitioning an empty dataset:

    first_block_schema = reduce_metadata_schema[0].schema
IndexError: list index out of range

[Copilot]

Pull Request Overview

This PR fixes the repartitioning behavior for empty datasets and refines the handling of block schemas during repartitioning.

• Added a new test to ensure empty datasets are repartitioned correctly with both shuffling options.
• Updated the block schema handling in the repartition task scheduler, including a default fallback for empty results.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
python/ray/data/tests/test_repartition_e2e.py	Added test case for empty dataset repartitioning with both shuffled and non-shuffled options.
python/ray/data/_internal/planner/exchange/split_repartition_task_scheduler.py	Refined block schema detection logic and improved error messaging for unknown block schemas.

Comments suppressed due to low confidence (1)

python/ray/data/tests/test_repartition_e2e.py:210

• [nitpick] Consider adding a docstring to this test function to clarify its purpose and expected behavior when repartitioning an empty dataset.

def test_repartition_empty_datasets(ray_start_regular_shared_2_cpus, shuffle):

[bveeramani]

Nit: Consider using the simplest inputs to make the intent clearer

Suggested change

	num_partitions = 5
	ds_empty = ray.data.range(100).filter(lambda row: False)
	num_partitions = 1
	ds_empty = ray.data.range(1).filter(lambda row: False)