GH-46025: [C++] Use ARROW_CUDA_EXPORT instead of ARROW_EXPORT for libarrow_cuda

[raulcd]

Rationale for this change

The code under arrow/gpu/ isn't included in libarrow.dll. We need to use different *_EXPORT macros for separated DLL.

What changes are included in this PR?

Create a new ARROW_CUDA_EXPORT macro to export cuda symbols.

Are these changes tested?

No new tests but should be tested via the cuda archery jobs.

Are there any user-facing changes?

No

• GitHub Issue: [C++] arrow/gpu/ should not reuse ARROW_EXPORT #46025

[raulcd]

@github-actions crossbow submit -g cuda

[github-actions]

Revision: 6a791cba771fa311205131724356a54a4307ae5e

Submitted crossbow builds: ursacomputing/crossbow @ actions-93a53001c3

Task	Status
test-cuda-cpp-ubuntu-22.04-cuda-11.7.1
test-cuda-python-ubuntu-22.04-cuda-11.7.1

[kou]

+1

But could you merge this after #46006 is merged and add Cflags.private: -DARROW_CUDA_STATIC to arrow-cuda.pc.in?

Related changes for adding Cflags.private: -DARROW_CUDA_STATIC to arrow-cuda.pc.in:

https://github.com/apache/arrow/pull/46006/files#diff-1bba462ab050e89360fd88110a689e85ee037749cea091a1848ab574381d3795R549-R554

https://github.com/apache/arrow/pull/46006/files#diff-7b9288301707cebf803cc90e8b4404277196b7041708f0219e466dcc9de8bc73R27-R28

[raulcd]

I'll wait until we merge #46006 (and rebase) to add the other relevant CMake changes, and will update arrow-cuda.pc.in with the:

Cflags:@ARROW_CUDA_PC_CFLAGS@
Cflags.private:@ARROW_CUDA_PC_CFLAGS_PRIVATE@

[raulcd]

@github-actions crossbow submit -g cuda

[github-actions]

Revision: 9aa1470

Submitted crossbow builds: ursacomputing/crossbow @ actions-2bef8745df

Task	Status
test-cuda-cpp-ubuntu-22.04-cuda-11.7.1
test-cuda-python-ubuntu-22.04-cuda-11.7.1

[raulcd]

@kou this should be ready for re-review

[kou]

+1

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit 770eaf4.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 3 possible false positives for unstable benchmarks that are known to sometimes produce them.