Kokkos-kernels: Fix #4630

Created by: mhoemmen

@trilinos/tpetra @trilinos/belos @vbrunini

Description

Modify Trilinos' snapshot of kokkos-kernels in order to fix #4630 (closed). We will add a separate kokkos-kernels PR with these changes.

Motivation and Context

Kokkos-kernels was not dispatching to cuBLAS for CudaUVMSpace Views. This made it slow. See discussion here: https://github.com/kokkos/kokkos-kernels/issues/397

Related Issues

• Closes #4630 (closed)
• Related to https://github.com/kokkos/kokkos-kernels/issues/396, https://github.com/kokkos/kokkos-kernels/issues/397

How Has This Been Tested?

TpetraCore tests pass in a CUDA static release build.