Update Kokkos policy kernels and timing

[marchdf]

Summary:

• Added CMake for Kokkos
• Added guarded/optional NVTX ranges around the main update phases for profiling
• Fused Kokkos update kernels for cu/cv/z/h, unew/vnew/pnew, and time smoothing. This removes launch overhead and memory traffic
• Added explicit Kokkos::fence() calls for accurate timing.
• Added per-phase timing and MFLOP reporting for loop 100, loop 200, and loop 300 (similar to other implementations)
• Used Kokkos::MemoryTraitsKokkos::Restrict on field views to help compiler opts

This results in 1.9X speedup (shallow_rangepolicy), 1.3X (shallow_rangepolicy_flat), 3.8X (shallow_mdrangepolicy), 1.2X (shallow_teampolicy)

This is one of several PRs stemming from a discussion with @sjsprecious and @johnmauff.