Replace the device model.

[trivialfis]

• Support leaf, QDM, inplace prediction for multi-target.
• Replace device model with tree-internal storage.
• Cleanup predict kernel.

[Copilot]

Pull Request Overview

This PR replaces the previous GPU device model with tree-internal storage and updates predictors to support leaf and multi-target predictions, alongside cleanup in prediction kernels.

• Replace uses of GetNodes().size() with Size() across code/tests.
• Introduce device-side tree views stored as cuda::std::variant and refactor GPU predictor kernels to consume them.
• Add device qualifiers and APIs in tree view for categorical splits; rework SHAP path extraction to use tree views.

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/cpp/tree/test_tree_model.cc	Update test to use RegTree::Size() API.
src/tree/tree_view.h	Return categories matrix by const reference; add device qualifiers; minor doc update.
src/tree/hist/evaluate_splits.h	Use RegTree::Size() for snode_ sizing.
src/predictor/gpu_predictor.cu	Major refactor: remove per-kernel MakeScalarTreeView; use device-resident variant of tree views; unify prediction kernels; add SHAP path extraction based on tree views.
src/predictor/cpu_predictor.cc	Use RegTree::Size() for tree_sizes_.

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[trivialfis]

cc @rongou

[Copilot]

Pull Request Overview

Copilot reviewed 14 out of 14 changed files in this pull request and generated 6 comments.

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}