Refactor Parallel_Reduce: Replace manual gather-reduce with MPI_Allreduce

[A-006]

Summary

This PR refactors the Parallel_Reduce class to use MPI_Allreduce directly for minimum and maximum operations, replacing the previous implementation that used MPI_Allgather followed by local reduction. This change simplifies the code, reduces memory overhead, and unifies the reduction interface.

Key Changes

• Refactored Parallel_Reduce Implementation:

  • Replaced gather_min_int_all with reduce_min_int.
  • Replaced gather_max/min_double_all with reduce_max/min_double.
  • Replaced gather_max/min_double_pool with reduce_max/min_double_pool.
  • The new methods utilize MPI_Allreduce with MPI_MIN/MPI_MAX for better efficiency.

• Updated Call Sites:

  • Updated all usages in source/source_estate/, source/source_pw/, source/source_lcao/, and other modules to match the new API.
  • Removed redundant #ifdef __MPI guards in caller functions, as the Parallel_Reduce wrappers now handle MPI conditional compilation internally.

• Cleanup:

  • Adjusted unit tests in parallel_reduce_test.cpp.

Impact

• Reduces code complexity.
• Potentially improves performance for reduction operations by leveraging optimized MPI implementations.
• Removes redundant MPI preprocessor directives from high-level code.