Simplify MPI trace merge logic and remove legacy guards

[marantic-amd]

Motivation

Remove conditional MPI compilation guards from perfetto trace merging
functionality to unify code paths. The merge logic now executes
unconditionally when ROCPROFSYS_PERFETTO_COMBINE_TRACES is enabled,
relying on MPI barrier and rank checks at runtime instead of
compile-time conditionals.

Technical Details

• Remove #if ROCPROFSYS_USE_MPI guards from get_perfetto_combined_traces()
  to allow configuration reading regardless of MPI availability
• Extract merge logic into dedicated merge_perfetto_files() function in
  filesystem_utils namespace for better organization
• Move merge script execution from cache_manager::post_process_bulk()
  into the new merge_perfetto_files() function
• Update mpi-perfetto-merge test to remove ROCPROFSYS_TRACE_LEGACY=ON
  requirement, as trace combining no longer depends on legacy mode

JIRA ID

ROCM-2141

Test Plan

• Existing and updated tests run.

Test Result

• Tests are passing on workstation machines

Submission Checklist

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