Fix WT KEDF CUDA reciprocal-space indexing

[Stardust0831]

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

No separate issue. This fixes the WT KEDF CUDA numerical regression introduced by #7448.

Unit Tests and/or Case Tests for my changes

• Existing coverage: tests/07_OFDFT/31_OF_KE_WT_GPU directly compares GPU WT energy, force, stress, and symmetry results with the CPU reference, so no new case or reference change is needed.
• Verification: CUDA build passed on Slurm job 748919. The single-rank 07_OFDFT/31_OF_KE_WT_GPU case passed all seven checks on Slurm job 748921: total energy -57.9338551428 eV, total force 0, and total stress 29.613417 kbar.
• Checks not run, with reason: no non-cubic or nspin > 1 WT CUDA integration case currently exists. The FFT dimension order was checked against the established FFT_CUDA and ig2ixyz_gpu layout.

What's changed?

• Apply each WT kernel value to its mapped full-box FFT coefficient through ig2ixyz_gpu, instead of assuming plane waves occupy the first npw FFT entries.
• Zero coefficients outside the selected plane-wave basis before the inverse FFT.
• Use the same (nx, ny, nz) cuFFT dimension order as ABACUS FFT_CUDA.

Governance Notes

• INPUT/docs changes: none. This restores the documented device gpu WT result and changes no parameter or interface.
• Core module impact: limited to the CUDA implementation of the WT KEDF convolution. CPU and other KEDF paths are unchanged.
• Exceptions requested: none.