[ci] Avoid fetching GTest with CMake.

[trivialfis]

Prevent failure: https://github.com/dmlc/xgboost/actions/runs/19546987883/job/55969434054

[hcho3]

Do you have a reason to believe that fetching gtest from Apt would lead to less flakiness than fetching it from GitHub?

[trivialfis]

No reason. But in general, having it handled by a dependency manager makes things easier. For instance, we can cache them in a docker image if needed. Also, we don't have to build from source everytime. I would like to replace the dmlc gtest whenever the chance arises.

[hcho3]

If so, let's use Conda to install gtest. This way, we can also use it for MacOS and Windows as well.

[Copilot]

Pull Request Overview

This PR modifies the gtest-cpu-nonomp CI workflow to use a conda-provided GTest package instead of fetching GTest through CMake via the dmlc-core submodule. This change addresses a failure in the CI pipeline where CMake's automatic GTest fetch was failing.

• Removed -DUSE_DMLC_GTEST=ON flag from the CMake build configuration
• Switched from apt-get installation to conda environment setup with miniforge
• Added proper bash shell configuration for conda environment activation

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
ops/pipeline/build-test-cpu-nonomp.sh	Removed the -DUSE_DMLC_GTEST=ON CMake flag to use system-provided GTest instead of dmlc-core's bundled version
.github/workflows/misc.yml	Replaced apt-get installation with miniforge-setup action, configured proper bash shell for conda, and added cpp_test conda environment setup

---

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