[Code scan] ABACUS HDF5 reader never loads k-points/eigenvalues due to misspelled key

This issue is part of a Codex global repository scan.

Problem:
The ABACUS HDF5 reader checks for `"kpionts"` but then reads `"kpoints"`. Normal HDF5 files containing `"kpoints"` and `"eigenvalues"` therefore skip band data loading silently.

Code reference:
https://github.com/deepmodeling/deeptb/blob/86c60c73996f0dd961c3138f2e88424382cb734e/dptb/data/dataset/_abacus_dataset.py#L40-L42

Impact:
ABACUS lazy/in-memory datasets can drop k-point and eigenvalue labels even when they are present, which can break or silently degrade band/eigenvalue training and evaluation.

Suggested fix:
Change the key check to `"kpoints" in data`, and consider adding a regression test that creates a minimal HDF5 file with `kpoints` and `eigenvalues`.