Implement device functions to simplify writing kernel code

[fwyzard]

Implement device functions to simplify writing kernel code:

• oncePerGrid(acc);
• oncePerBlock(acc);
• uniformElements(acc, size) and uniformElements(acc, begin, end);
• uniformGroups(acc, size) and uniformGroupElements(acc, group, size);
• independentGroups(acc, groups) and independentGroupElements(acc, group_size);
• variations of these functions along different dimensions.

Implement tests for the most common functions.

[fwyzard]

let's see if the 23rd time's a charm...

[SimeonEhrig]

I check your code locally. The problem is the AccCpuThreads backend. On our workstation with 32 Core Epyc and Clang 17 debug build, the test already runs 100s. I think the annotation of the thread sanitizer also add a big overhead and the CI runners are weaker than our workstation.

I suggest to disable the tests with std::thread backend or if it make sense to test the thread backend, decrease the problem size.

[fwyzard]

I suggest to disable the tests with std::thread backend or if it make sense to test the thread backend, decrease the problem size.

I agree... I've tried reducing the block size in all tests to 32 threads (or elements) per block.

[fwyzard]

Closed in favour of https://github.com/alpaka-group/alpaka/pull/2369 that includes further fixes and clean up.