fix: fixed incorrect use of static assert when checking device warp size

[Naraenda]

Closes https://github.com/ROCm/rocPRIM/issues/542.

This change introduces a ROCPRIM_DETAIL_DEVICE_STATIC_ASSERT macro (internal detail macro and is not public API). That only asserts when compiling for device, thus skipping the assert when compiling on host. Extra tests have been added in test/rocprim/test_block_load_store.kernels.hpp. Some extra mechanisms have been added to compile dummy kernels instead of invalid ones for invalid configurations. Similarly the test code does not invoke test cases with dummy kernels and instead skips them.

Code has already been internally reviewed, but please do go over it before merging.