- Add new BatchedGemm function to the KokkosBatched namespace.
  This function selects how to invoke the SerialGemm routine based
  on the template arguments, target architecture, and Kokkos backend.
  The invocation selection was informed by performance data collected
  on V100, intel, and A64FX using the batched gemm performance tests.
  - Add unit tests for the BatchedGemm function.

  - Move contents of src/common/KokkosKernels_Macros.hpp
  to src/KokkosKernels_Macros.hpp.
  - Update KokkosSparse_CrsMatrix.hpp and KokkosBatched_Gemm_Decl.hpp
  to use unused macro.