fix:add OpenMP parallelization to BPCG CPU kernel band loops (assisted with Deepseek V4)

[Missing-Hex]

Linked Issue

Add OpenMP parallelization to BPCG CPU kernels for improved performance

Unit Tests and/or Case Tests for my changes

• Existing unit tests in source/source_hsolver/kernels/test/ cover the BPCG functionality
• Verified numerical correctness by running existing test cases with both single-threaded and multi-threaded configurations
• Performance benchmarking shows significant speedup with multi-threaded execution

What's changed?

• Added OpenMP parallelization directives (#pragma omp parallel for) to the outer band loops in line_minimize_with_block_op<CPU> and calc_grad_with_block_op<CPU> in bpcg_kernel_op.cpp
• Added #pragma omp critical sections to protect MPI reduction calls (Parallel_Reduce::reduce_pool) ensuring thread safety
• All OpenMP directives are wrapped with #ifdef _OPENMP for conditional compilation, maintaining backward compatibility with non-OpenMP builds
• The parallelization targets band-level data parallelism where each band's computation is independent
• Expected performance improvement scales with the number of bands and available CPU cores

Any changes of core modules? (ignore if not applicable)

• Modified HSolver module: Enhanced bpcg_kernel_op.cpp with OpenMP parallelization for CPU kernels
• The changes affect line_minimize_with_block_op and calc_grad_with_block_op structures which are critical for BPCG diagonalization performance
• No changes to class interfaces or virtual functions - modifications are purely implementation-level optimizations
• GPU code paths remain unchanged; this optimization only affects CPU execution