[userchba]

fix(gpu): support LibDDLA 0.0.6 public API

[Stardust0831]

Thanks for the update! I tested this PR at df004e5 with the official DDLA v0.0.6 release on NVIDIA V100 16 GB GPUs, using GCC 13.3, CUDA 12.9 and Open MPI 5.0.10.

The following checks passed:

• CPU/CUDA builds and all 38 symmetry regression groups with both 1 and 4 OpenMP threads.
• GPU contractions and accumulation for all four scalar types, plus CPU/GPU numerical comparisons for the EXX Hamiltonian, forces, stress and density updates.
• A clean ABACUS build followed by 20 HSE runs and 16 numerical comparisons, covering real/complex tensors, one/two MPI ranks and GPUs, and symmetry enabled/disabled. Eight runs exercised GPU EXX contractions.

No additional LibRI patches were needed, and existing reference values and tolerances were preserved. These tests cover single-node CPU/CUDA execution; HIP and multi-node execution were not tested.

From the ABACUS integration side, these tests did not reveal a blocker. If there are no other review concerns, could you consider merging this PR and publishing a LibRI release containing #15, #16 and #19? ABACUS #8169 has been adapted and tested, and remains a draft pending a compatible official LibRI release so that downstream users have a clearly specified dependency version.