Refactor: Replace SFINAE by if constexpr for create_mirror* functions

[pzehner]

This PR is a refactor without side effects.

The idea is to simplify the code of the create_mirror* functions by applying techniques of the C++ 17 standard. A refactor with the similar approach was drafted by @masterleinad in #5533 about two years ago, but never merged.

The multiple overloaded functions of Impl::create_mirror, Impl::create_mirror_view, and Impl::create_mirror_view_and_copy that use SFINAE are replaced by one function that uses if constexpr instead (for Kokkos views, offset views, dynamic views and dynamic rank views). Eventually, the return type of these functions is replaced by auto, and decided depending on which branch of the if constexpr is used.

The logic of the public create_mirror* functions (which is sometimes redundant with the logic of the private functions) is not altered in this PR and will be assessed in an upcoming one.

This PR is part of a of a larger effort which aims to assess #6842.

We successfully built and tested the refactored code for the following configurations:

Compiler	Version	Min. ver.	Backend	Version	Min. ver.
Clang	14.0.0	8.0.0	Serial
GCC	8.4.0	8.2.0	Serial
GCC	8.4.0	8.2.0	OpenMP
GCC	8.4.0	8.2.0	CUDA	11.4	11.0
Intel Classic	2023.2.1	19.0.5	OpenMP
Intel LLVM	2023.2.1	2021.1.1	OpenMP
Intel LLVM	2024.0.2	2022.0.0	SYCL
HIP	5.2.0	5.2.0	ROCM

where "Compiler" is the used compiler, next "Version" is the version of the compiler, next "Min. Vers." is the minimum version of the compiler for Kokkos 4.3.0 according to the documentation, "Backend" is the name of the backend, next "Version" is the version for the backend if necessary, and next "Min. Vers." is the minimum version of the backend supported.

Unfortunately, we could not test any earlier version of the compilers, nor of the backends, on our test machines. Also, please note that some configurations could not be used:

• Clang with OpenMP or CUDA;
• MSVC; and
• NVHPC.

Finally, the SYCL configuration was built, but not tested.

[masterleinad]

You'll have to fix indentation.

[pzehner]

The code style was fixed using clang-format.

[masterleinad]

Looks OK to me but we should test with Trilinos/Sacado/Stokhos just to make sure this doesn't break their customizations.

[masterleinad]

Can you please rebase on top of develop (or merge) to avoid running with broken configurations?

[pzehner]

Merging done. For merging, can you squash the commits, as the history of this branch is quite messy?

[masterleinad]

Merging done. For merging, can you squash the commits, as the history of this branch is quite messy?

Yes, we normally squash commits before merging.

[pzehner]

I think I merged the latest version of develop, but the same build errors not related to the PR are still showing up.

[pzehner]

we should test with Trilinos/Sacado/Stokhos just to make sure this doesn't break their customizations.

Could you recommend any build script for Trilinos, in order to test it?

[masterleinad]

I think I merged the latest version of develop, but the same build errors not related to the PR are still showing up.

Yes, it seems an update in our CI containers is responsible for the build failures. #6969 fixes one of them.

[crtrott]

Retest this please!

[pzehner]

@crtrott Could you validate this PR? It seems the failing jobs are due to external factors.