GH-45510: [CI][C++] Fix LLVM APT repository preparation on Debian

[kou]

Rationale for this change

The existing LLVM APT repository preparation is broken. For example, it uses unavailable ${available_llvm}.

What changes are included in this PR?

• Use .asc for armored key
• Use deb822 format for APT source: https://manpages.debian.org/bookworm/dpkg-dev/deb822.5.en.html

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

• GitHub Issue: [CI][C++] test-debian-12-cpp-amd64 failed #45510

[kou]

@github-actions crossbow submit test-debian-12-*

[github-actions]

⚠️ GitHub issue #45510 has been automatically assigned in GitHub to PR creator.

[github-actions]

Revision: 23cac6e

Submitted crossbow builds: ursacomputing/crossbow @ actions-81ac5f56f9

Task	Status
test-debian-12-cpp-amd64
test-debian-12-cpp-i386
test-debian-12-docs
test-debian-12-python-3-amd64
test-debian-12-python-3-i386

[raulcd]

The failure on debian-12-docs doesn't seem related to this change:

2025-02-12T08:43:06.1451336Z -- Unable to find cuda_runtime.h in "/usr/lib/include" for CUDAToolkit_INCLUDE_DIR.
2025-02-12T08:43:06.1452770Z -- Unable to find cublas_v2.h in either "" or "/math_libs/include"
2025-02-12T08:43:06.1476771Z CMake Error at /usr/share/cmake-3.25/Modules/FindPackageHandleStandardArgs.cmake:230 (message):
2025-02-12T08:43:06.1477685Z   Could NOT find CUDAToolkit (missing: CUDAToolkit_INCLUDE_DIR) (found
2025-02-12T08:43:06.1478231Z   version "11.8.89")
2025-02-12T08:43:06.1478583Z Call Stack (most recent call first):
2025-02-12T08:43:06.1479324Z   /usr/share/cmake-3.25/Modules/FindPackageHandleStandardArgs.cmake:600 (_FPHSA_FAILURE_MESSAGE)
2025-02-12T08:43:06.1480405Z   /usr/share/cmake-3.25/Modules/FindCUDAToolkit.cmake:869 (find_package_handle_standard_args)
2025-02-12T08:43:06.1481189Z   src/arrow/gpu/CMakeLists.txt:44 (find_package)

I can see nvidia-cuda-toolkit being installed successfully. We probably want to open that as a separate issue and merge this.

[kou]

It makes sense. Let's do it.

[raulcd]

ok, I've opened the issue mentioned above. I am merging this.
Thanks @kou

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit c2432af.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 10 possible false positives for unstable benchmarks that are known to sometimes produce them.