Trilinos nightly failure, Hip builds on MI210, hip.reduce_device_view_mdrange_policy fails in builds with MPI

[ndellingwood]

Describe the bug

Nightly Trilinos Hip builds are failing in the hip.reduce_device_view_mdrange_policy unit test when the job is run with 4 mpi processes:

21:34:12 [ RUN      ] hip_managed.view_allocation_large_rank
21:34:12 Memory access fault by GPU node-2 (Agent handle: 0x5f27b20) on address 0x7fa871800000. Reason: Unknown.
21:34:12 --------------------------------------------------------------------------
21:34:12 Primary job  terminated normally, but 1 process returned
21:34:12 a non-zero exit code. Per user-direction, the job has been aborted.
21:34:12 --------------------------------------------------------------------------
21:34:12 --------------------------------------------------------------------------
21:34:12 mpirun noticed that process rank 0 with PID 3958738 on node lean1 exited on signal 6 (Aborted).
21:34:12 --------------------------------------------------------------------------

The job is set to rerun failed tests when failures occur and the test failed in both the initial run and rerun

[ndellingwood]

This test passed in the most recent run of nightlies, we should keep this issue open for tracking occurrences