WIth these changes MPI, off and hacking the Kokkos::Experiment::HIP::memory_space
typedef to be Kokkos::Experimental::HIPHostPinnedSpace we get these failures:

124/124 Test #124: TpetraCore_TsqrAdaptor ..........................................   Passed    0.55 sec
95% tests passed, 5 tests failed out of 124
Label Time Summary:
Tpetra    = 147.76 sec*proc (124 tests)
Total Test time (real) = 147.96 sec
The following tests FAILED:
         19 - TpetraCore_idot (Subprocess aborted)
         83 - TpetraCore_MatrixMatrix_UnitTests (NUMERICAL)
        121 - TpetraCore_RowMatrixTransposer_test (Subprocess aborted)
        122 - TpetraCore_RowMatrixTransposer_UnitTests (Failed)
        123 - TpetraCore_CrsMatrix_transpose_sortedRows (Failed)

This IS running on the AMD GPU ...

- Added explicit instantiation file for FixedHashTable
- Use HIP/HIPHostPinnedSpace as the KokkosHIPWrapperNode in Tpetra only
  (no kokkos changes)
- Fix device types in several places to work when the desired device
  memory space isn't the default of the execution space
- Disabled 3 tests that crash AMD GPUs