Remove the deprecated gpu_hist.

[trivialfis]

No description provided.

[Copilot]

Pull Request Overview

This PR removes the deprecated gpu_hist tree method across the codebase, replacing its usage with hist or approx, and cleans up related GPU configuration hooks.

• Replace all references to gpu_hist in tests, docs, and code with supported methods (hist or approx).
• Remove UseGPU overrides and ConfigureGpuId logic now that gpu_hist is gone.
• Update Spark, Python, Scala, and C++ layers to drop unsupported gpu_hist handling.

Reviewed Changes

Copilot reviewed 17 out of 17 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/test_distributed/test_with_spark/test_spark_local.py	Replaced gpu_hist with hist/approx in Spark tests
tests/python-gpu/test_gpu_pickling.py	Refactored pickling test into a helper, added type hint
tests/cpp/gbm/test_gbtree.cc	Updated mapping table and expectations for approx
src/learner.cc	Removed deprecated ConfigureGpuId calls
src/gbm/gbtree.h	Dropped kGPUHist enum and registration
src/gbm/gbtree.cc	Removed TreeMethod::kGPUHist updater and fallback
src/gbm/gblinear.cc	Removed GPU support checks and UseGPU override
src/context.cc	Deleted deprecated ConfigureGpuId method
python-package/xgboost/spark/core.py	Simplified _run_on_gpu logic, dropped gpu_hist case
python-package/xgboost/sklearn.py	Removed gpu_hist from _can_use_qdm
jvm-packages/.../XGBoostEstimatorSuite.scala	Deleted tests for gpu_hist in Scala suite
jvm-packages/.../TreeBoosterParams.scala	Removed gpu_hist from supported methods
jvm-packages/.../XGBoostEstimator.scala	Simplified runOnGpu logic
jvm-packages/.../GpuXGBoostPluginSuite.scala	Removed gpu_hist validation tests
jvm-packages/.../GpuXGBoostPlugin.scala	Adjusted require check to only depend on device
include/xgboost/gbm.h	Removed virtual UseGPU declaration
include/xgboost/context.h	Deleted deprecated GPU configuration doc and method

Comments suppressed due to low confidence (2)

python-package/xgboost/spark/core.py:504

• [nitpick] Update this docstring to clarify that GPU execution now depends solely on device or the use_gpu flag, since tree_method no longer triggers a GPU path.

        """If train or transform on the gpu according to the parameters"""

tests/test_distributed/test_with_spark/test_spark_local.py:930

• Add an assertion after validation to verify that clf._run_on_gpu() returns True when device='cuda' and tree_method='approx', ensuring the new approximate GPU path is tested.

        clf._validate_params()