Unify RT-TDDFT EXX with the general LCAO H(R) path

[ESROAMER]

What's changed?

This PR unifies the RT-TDDFT hybrid-functional EXX workflow with the general LCAO H(R) Hamiltonian path.

Previously, RT-TDDFT used a dedicated k-space EXX path (Add_Hexx_Type::k and add_Hexx_td), while the ordinary LCAO workflow accumulated EXX in H(R). This resulted in separate
Hamiltonian assembly and gauge-phase handling for RT-TDDFT.

The new implementation uses the same H(R)-based EXX path for both ordinary LCAO calculations and RT-TDDFT:

• EXX is accumulated into the complete H(R) together with the local and non-local Hamiltonian terms.
• The final generic OperatorLCAO folding step constructs H(k) and applies the time-dependent hybrid-gauge phase consistently to all Hamiltonian contributions.
• The RT-TDDFT-specific k-space EXX reconstruction and add_Hexx_td interface are removed.
• OperatorEXX::contributeHk() is retained only for dedicated Add_Hexx_Type::k consumers such as write_vxc and RDMFT.
• Hybrid RT-TDDFT selects a complex H(R) representation even for collinear spin cases.
• The complex Veff path is adapted so that Gint contributions are correctly accumulated into complex H(R).
• Duplicate EXX reconstruction in the current/H(R) assembly path is removed.
• RT-PBE0 reference results are updated to the validated H(R)-based implementation.

The main purpose of this change is to eliminate the divergent RT-TDDFT EXX path and make RT-TDDFT use the same H(R)-based Hamiltonian construction as the general LCAO workflow.

[AsTonyshment]

Nice work overall! Unifying the RT-TDDFT EXX path with the general LCAO H(R) workflow makes the Hamiltonian construction much cleaner. I had also noticed the fragmented paths summarized in the diagram, but had not found time to address them. I left one inline concern about the interaction between Cell_Nearest and the hybrid-gauge phase; otherwise, I think this is a very good direction.

[AsTonyshment]

I may be missing something about how the LibRI cell indices are defined, but I think applying Cell_Nearest here may lose information needed by the hybrid-gauge phase. BvK-equivalent translations have the same Bloch phase on the sampled k-point mesh, but their finite-field phases, exp(i A(t)·R), are generally different 🤔