Support skewed in-plane SPM cells

[cpignedoli]

Summary

This draft adds initial support for skewed in-plane CP2K SPM cells in the WFN gridding path.

The implementation keeps the existing orthogonal-cell behavior intact while introducing full cell-vector handling for slab-like cells where A and B lie in the xy plane and C is along z. More general tilted/triclinic cells are deliberately rejected for now instead of being treated incorrectly.

This PR is stacked on top of #20 (codex/numpy2-compat) and should be retargeted to main after #20 is merged.

Changes

• Parse CP2K A, B, and C cell vectors without requiring a rectangular cell.
• Store full cell vectors alongside the existing cell-vector lengths used by older code paths.
• Convert atom positions through fractional/grid coordinates before placing local orbital grids.
• Evaluate radial functions and solid harmonics in Cartesian coordinates on skewed grids.
• Use the in-plane reciprocal metric for FFT extrapolation instead of assuming kx**2 + ky**2.
• Write CUBE files with the full evaluation-cell matrix and integrate densities using the voxel determinant.
• Convert p-tip finite-difference derivatives from lattice-axis derivatives to Cartesian in-plane gradients.
• Save skewed-cell metadata in STM/orbital .npz outputs.
• Add examples/validate_skewed_cell_support.py for lightweight manual validation with bundled benzene data.

Validation

• python -m py_compile cp2k_spm_tools/cp2k_grid_orbitals.py cp2k_spm_tools/cp2k_stm_sts.py examples/validate_skewed_cell_support.py
• git diff --check
• PYTHONPATH=/home/jovyan/opt/cp2k-spm-tools python examples/validate_skewed_cell_support.py
• pre-commit run --files cp2k_spm_tools/cp2k_grid_orbitals.py cp2k_spm_tools/cp2k_stm_sts.py examples/validate_skewed_cell_support.py

Additional local check: the standard orthogonal benzene CUBE output was compared against the base NumPy-2 branch and matched exactly for HOMO and LUMO.

Still To Test

• Real non-90-degree CP2K/SPM calculations from production-like inputs.
• Plotting behavior for skewed STM/orbital maps.
• HRSTM-specific paths, which still contain additional rectangular-grid assumptions and are not claimed as fixed here.

[cpignedoli]

Follow-up fix pushed in 21b778b: a real AiiDA STM submission exposed that the ionization-potential diagnostic assumed the HOMO is always present in the selected STM orbital window. That is not guaranteed for finite energy-window runs, so the diagnostic now skips itself cleanly when the HOMO is outside the selected window instead of crashing the STM CalcJob.

Extra validation after the fix:

• python -m py_compile cp2k_spm_tools/cli/stm_sts_from_wfn.py
• git diff --check
• pre-commit run --files cp2k_spm_tools/cli/stm_sts_from_wfn.py
• Replayed the failing local STM command against the generated CP2K parent folder; it now completes and writes stm.npz.

[cpignedoli]

Added a follow-up validation for the p-tip path.

What changed:

• Factored the in-plane p-tip gradient norm into STM.in_plane_gradient_squared() without changing the formula.
• Added an analytic skew-grid check in examples/validate_skewed_cell_support.py using a linear field with known Cartesian gradient.
• The check verifies that the skewed-grid axis derivatives recover the correct grad_x**2 + grad_y**2 p-tip contribution.

Validation run:

• python examples/validate_skewed_cell_support.py
• python -m py_compile cp2k_spm_tools/cp2k_stm_sts.py examples/validate_skewed_cell_support.py
• pre-commit run --files cp2k_spm_tools/cp2k_stm_sts.py examples/validate_skewed_cell_support.py

Prepared with assistance from Codex.