Fix/gpu elecstate pw op spin stride

[chengleizheng]

Reminder

• I have read AGENTS.md and docs/developers_guide/agent_governance.md.
• I have linked an issue or explained why this PR does not need one.
• I have added adequate unit tests and/or case tests, or explained why not.
• I have listed the exact verification commands run and their results.
• I have described user-visible behavior changes, including INPUT parameter changes.
• I have explained core-module impact for ESolver, HSolver, ElecState, Hamilt, Operator, Psi, or other source/ changes.
• I have requested any needed governance exception below.

Linked Issue

Fix #7773

Case Tests for my changes

• Commands run:
  • Case test tests/01_PW/007_PW_UPF201_USPP_Fe (PW SCF, USPP, nspin=2, ecutwfc=8, ecutrho=60 — the dense FFT grid is 4× larger per dimension than the smooth grid, which is exactly the configuration that exposed the bug).
  • CPU reference (single process): abacus on 007 → FINAL_ETOT_IS = -673.8349347015964668 eV.
  • GPU (CUDA build, single process): same case → FINAL_ETOT_IS = -673.8349347005279242 eV (GPU kernels active, e.g. PW_Basis_K recip_to_real gpu in timing).
• Result summary:
  • Post-fix GPU etot agrees with CPU to within 1.1×10⁻⁹ eV (Δ = |−673.8349347005279242 − (−673.8349347015964668)|).

What's changed?

• Bug: the GPU kernel elecstate_pw in source/source_estate/kernels/cuda/elecstate_op.cu wrote charge density with rho[spin * basis->nrxx + idx], using the wavefunction grid stride (basis->nrxx, the smooth grid). But the rho array is laid out per spin on the density grid (charge->nrxx, the dense grid). For USPP (double grid) the two differ (e.g. 9³ vs 12³ on 007), so the spin-down component was written at a wrong offset and effectively lost → spin-down charge density missing → etot off by 6.07 eV on GPU.
• Fix: the kernels now take an explicit rho_stride parameter, and every caller passes charge->nrxx:
  • elecstate_op.cu / elecstate_op.h / elecstate_op.cpp: kernel and operator() signatures extended with const int& rho_stride (CPU ignores it; its per-spin pointer indexing was already correct).
  • elecstate_pw.cpp: rho accumulation site (rhoBandK) and kin_r site now pass charge->nrxx.
  • elecstate_pw_cal_tau.cpp: passes charge->nrxx for the kinetic energy density.
  • read_wf2rho_pw.cpp (2 sites) and sto_iter.cpp: pass charge->nrxx.

Governance Notes

• INPUT/docs changes: none.
• Core module impact: ElecState GPU path only — the CUDA kernel interface in source_estate/kernels is extended with one parameter; all GPU call sites within this repo are updated in this PR. CPU path behavior is unchanged.
• Exceptions requested: none.