[FEA] Add getDeviceCapability function

Spawned from this cuML PR comment: https://github.com/rapidsai/cuml/pull/3058#discussion_r518439447

This is a simple utility method that would be good to have in raft.

#include <utility>  // pair

/** helper method to get the compute capability version numbers */
inline std::pair<int, int> getDeviceCapability() {
  int devId;
  CUDA_CHECK(cudaGetDevice(&devId));
  int major, minor;
  CUDA_CHECK(
    cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, devId));
  CUDA_CHECK(
    cudaDeviceGetAttribute(&minor, cudaDevAttrComputeCapabilityMinor, devId));
  return std::make_pair(major, minor);
}

Start with the linked cuML pull request discussion and review the surrounding RAFT CUDA utility conventions. Add the supplied getDeviceCapability helper at the appropriate existing utility entry point, returning the current device's major and minor compute capability with the shown CUDA error checks. Done means the helper is available and relevant existing CUDA tests pass.