MPI Fortran push-pop count fix for extra MAIN subroutine

[sputhala-amd]

Motivation

Fix the CI failure for MPI Fortran tests due to MAIN subroutine not getting popped from tracing

Technical Details

Fix CI assert failure by ensuring extra Fortran MAIN subroutine is popped correctly, preventing push/pop trace count mismatch.
The failure was happening when built with ROCPROFSYS_USE_MPI=ON and not with ROCPROFSYS_USE_MPI_HEADERS=ON. If the finalization call is preceded with an additional pop trace for MPI_Finalize/PMPI_Finalize then the MAIN is popped correctly for the full MPI case with ROCPROFSYS_USE_MPI=ON..

Test Plan

Manual tests were performed with the fix.

Test Result

The testing confirms the fix works.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[dgaliffiAMD]

Are there additional Fortran tests that we should enable after your fix?

[sputhala-amd]

1. Add MPI_Finalize/PMPI_Finalize in the traces.
2. Check for Fortran additional subroutines during finalization and pop the count for trace explicitly as we do for main executable.

[The instrumentation bundle for the main binary and the additional main subroutine (in case of Fortran) are created during library startup inside dl.cpp, these will be popped from the trace later during finalization.]

[sputhala-amd]

Are there additional Fortran tests that we should enable after your fix?

@dgaliffiAMD , Kian has a branch with the MPI Fortran specific tests #349. I have requested Kian to test with this fix and confirm.

[kcossett-amd]

See #1172 for ctests (labelled mpi-fortran)

[sputhala-amd]

This Fix is not needed as another PR fixes this issue.