feat: add ray repartition pipeline

[macroguo-ghy]

Summary

• add a Ray-only repartition pipeline for dataset-level block repartitioning
• register the pipeline and list it in config_all.yaml and operator docs
• add unit coverage for repartition options, local-executor failure, and load_ops registration

Validation

• pre-commit run --files data_juicer/ops/pipeline/repartition_pipeline.py data_juicer/ops/pipeline/init.py data_juicer/ops/mapper/init.py data_juicer/config/config_all.yaml docs/Operators.md docs/operators/pipeline/repartition_pipeline.md tests/ops/pipeline/test_repartition_pipeline.py
• python3 -m pytest tests/ops/pipeline/test_repartition_pipeline.py
• python3 -m py_compile data_juicer/ops/pipeline/repartition_pipeline.py tests/ops/pipeline/test_repartition_pipeline.py && git diff --check
• load_ops registration smoke check

[gemini-code-assist]

Code Review

This pull request introduces the repartition_mapper operator, which allows for repartitioning Ray Datasets into a target number of blocks. The implementation includes the operator logic, configuration updates, documentation, and unit tests. Review feedback consistently points out that since the operator inherits from the Pipeline class and performs dataset-level transformations, it should be renamed to RepartitionPipeline and reclassified from a mapper to a pipeline across the codebase and documentation to maintain architectural consistency.

[cmgzn]

This is a very useful new op, thanks! The implementation looks good to me. One minor suggestion: since it is Ray-only, maybe rename it to ray_repartition_pipeline for consistency with other Ray-only ops. If you prefer, we can also make this adjustment on our side.

[macroguo-ghy]

Thanks for the suggestion! Renamed it to ray_repartition_pipeline and updated the code, config, docs, and tests accordingly.