Use context for thrust caching policy.

[trivialfis]

Replace all existing use of the free function.

[Copilot]

Pull request overview

Updates GPU-side Thrust usage to rely on the CUDA Context/CUDAContext for the caching execution policy (instead of a free helper), ensuring Thrust operations consistently use the context’s allocator and stream.

Changes:

• Remove the dh::CachingThrustPolicy() free function from device_helpers.cuh.
• Update FeatureInteractionConstraintDevice::Split to accept a Context and use ctx->CUDACtx()->CTP() for the Thrust fill.
• Update call sites (GPU hist updater + CUDA tests) to pass a CUDA context and use CTP() where needed.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/cpp/tree/test_constraints.cu	Create CUDA context in GPU constraint tests and pass it into Split/QueryNode; minor formatting cleanups.
tests/cpp/common/test_device_vector.cu	Use ctx.CUDACtx()->CTP() for Thrust algorithms in tests; add required includes/helpers.
src/tree/updater_gpu_hist.cu	Pass ctx_ into interaction-constraint Split during tree expansion.
src/tree/constraints.cুহ	Update FeatureInteractionConstraintDevice::Split signature to accept Context const*.
src/tree/constraints.cu	Use ctx->CUDACtx()->CTP() for the buffer clear after split.
src/common/device_helpers.cuh	Remove CachingThrustPolicy() helper.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.