Subgroup tests - sizeof() for builtin return values is not tested

The sizeof() result for return values of subgroup builtin functions is generally not tested.

This should be done with dedicated kernels rather than as part of the testing of subgroup operations.

What is the purpose of the sizeof() calls in the cl_khr_subgroup_clustered_reduce tests? Are these intended to test the size of the built functions' return values? If so, this testing should be rewritten as described above.

Start by inspecting the cl_khr_subgroup_clustered_reduce tests and their existing sizeof() calls. Determine whether those calls cover builtin return-value sizes, then add dedicated kernels for that coverage and verify the subgroup tests still exercise the intended cases.

Written by the indexing model from the issue text.