[enc] Throw error when DMatrix is empty.

[trivialfis]

No description provided.

[Copilot]

Pull Request Overview

This PR adds error handling to throw an error when a DMatrix is empty in a distributed setting with categorical features. The change prevents issues that could arise when some workers have empty input while others have categorical data, which would make category inference impossible.

• Added validation to check for empty DMatrix when categorical features are present in distributed settings
• Refactored Dask utilities to move version checking functions to a separate module
• Updated tests to expect the new validation error rather than proceeding with empty data

Reviewed Changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/data/cat_container.cc	Implements the core validation logic that throws an error when any worker has empty input with categorical features
src/data/cat_container.h	Adds function declaration for the new validation function
src/data/*.cc (multiple)	Integrates the validation check across different DMatrix implementations
tests/test_distributed/test_gpu_with_dask/test_gpu_with_dask.py	Updates test to expect ValueError when training with empty DMatrix
python-package/xgboost/dask/utils.py	Moves Dask version checking utilities to separate module
python-package/xgboost/dask/init.py	Imports version checking functions from the new utils location

[trivialfis]

cc @rongou .