Restore CP2K grid integration parallel baseline

[guanLeTea]

Promotes cp2k_grid_integrate to a level 3 represenation by restoring the parallel baseline from the upstream
Changes:
-Relocated cp2k_grid_integrate to "hpcagent_bench" canonical benchmark location
-Restore task-level OpenMP parallelism
-Preserve SIMD vectorization
-Updated YAML
-Updated port tests

Validation (local)
-Port test passes
-e2 runs
-Numerical validation with the reference file

Promotes cp2k_density_matrix_trs4 to a level 4 represnation by restoring meaningful production-style parallism
Changes:

• Relocated cp2k_density_matrix_trs4 to the canonical "hpcagent_bench" benchmark location
• Added the canonical _fp64 Fortran ABI wrapper derived from the manifest binding
• Restored DBCSR-style block-row OpenMP parallelism into standalone sparse matrix multiplication
• Preserved serial execution for numerically sensitive TRS4 logic (trace, gamma, convergence, branch selection)
• Updated YAML
• Updated and extended port tests

Validation (local):

• BenchSpec discovery succeeds
• Port tests pass
• End-to-end numerical tests pass
• Numerical validation against the NumPy/Fortran reference
• Bitwise-identical results with OMP_NUM_THREADS=1, 2, and 4
• OpenMP-enabled native reference verified