PR NVIDIA/cub#218 fixes this CUB's radix sort. We should:

• Check whether Thrust's other backends handle this case correctly.
• Provide a guarantee of this in the stable_sort documentation.
• Add regression tests to enforce this on all backends.

The functions under metrics and score in src_prims can be semantically categorized under a common name (whether that can be named 'metrics' or 'scores' is open for discussion).

It is exposed in cuml under a common rubric as [metrics](https://github.

An assignment within an if-statement introduced in [bac2060e991582063775d34eaeb40822810e4349] triggers a warning with Clang:

In file included from /thrust/cub/block/../iterator/cache_modified_input_iterator.cuh:42:
/thrust/cub/block/../iterator/../util_device.cuh:323:35: warning: using