Introduce WFN-based extrapolation for NAO calculations

[Growl1234]

ABACUS currently mainly relies on charge-density extrapolation between ionic steps. This is robust, but it only extrapolates the density and does not preserve or predict the occupied orbital subspace.

For NAO calculations, a WFN-based extrapolation framework may provide a better initial electronic state for MD and structural relaxation. The general workflow is:

WFN history
  -> predict initial WFN
  -> reorthonormalize with current S / S(k)
  -> rebuild DMK / DMR / rho
  -> enter SCF

For k-point calculations the case will be more complicated. Besides complex wavefunctions and per-k S(k), the implementation should handle local/global k-point indexing, phase consistency under PBC wrapping, subspace alignment, and safe reorthonormalization with the current overlap matrix.

TODO

• Phase 1: Add WFN-history infrastructure through use_prev_wf

  • Add wfc_extrap input option.
  • Add source_lcao/module_extrap.
  • Add owned WFN snapshot/history buffer.
  • Implement Gamma-only use_prev_wf.
  • Rebuild DMK / DMR / rho from the predicted WFN.

• Phase 1.5: Optimize and harden the WFN-extrapolation backend (necessary before adding tests and expanding to advanced features)

• Phase 2: Enable use_prev_wf for k-points

  • Support complex WFN snapshots.
  • Handle per-k S(k).
  • Add phase/convention correction.
  • Support local/global k-point mapping.

• Phase 3: Add ASPC WFN extrapolation

  • Add multi-step WFN history.
  • Add ASPC coefficients.
  • Implement projected-WFN construction.

• Phase 4: Add GExt_PROJ WFN extrapolation

  • Add overlap-difference fitting.
  • Add k-weighted inner products.
  • Add safeguards for ill-conditioned fitting.