refactor(pw): optimize bspline structure factor grid

[goodchong]

Summary

• Populate only the local FFT z slab in bspline structure factor setup
• Remove the full nxyz staging array and zpiece_to_all scatter path
• Use vector-managed temporary buffers for bspline grids and coefficients

Tests

• git diff --check
• cmake --build /tmp/abacus-structure-factor-compile --target module_pwdft -j2
• /tmp/abacus-structure-factor-compile/abacus_pw_ser in tests/01_PW/032_PW_15_CF_CS_bspline