Fix TPL_LIBRARY_SUFFIXES for 32-bit build

[masterleinad]

For 32-bit builds, we should search for TPL libraries in lib32 instead of lib64. This should fix the last failing GitHub CI build.

The openmp.parallel_scan_with_reducers test is failing with

 RUN      ] openmp.parallel_scan_with_reducers
/__w/kokkos/kokkos/core/unit_test/TestTeamVector.hpp:1010: Failure
Expected equality of these values:
  host_outputs(i)
    Which is: 1.0302e-12
  expected(i)
    Which is: 1.0302e-12
differ at index 103
/__w/kokkos/kokkos/core/unit_test/TestTeamVector.hpp:1010: Failure
Expected equality of these values:
  host_outputs(i)
    Which is: 1.0302e-12
  expected(i)
    Which is: 1.0302e-12
differ at index 102

and I decided to just skip since we have multiple other tests disabled for this build already (and it's not related to the build issue fixed here).

[masterleinad]

The 32-bit build is passing.

[dalg24]

Ignoring unrelated runtime issues in one HIP build and one CUDA build that timed out