Refactor: read EXX ABFS/JLE file lists via UnitCell, fully decouple source_cell (#7598)

[Critsium-xy]

Background

Related to #7598 (source_cell should not depend back on higher layers).

After the earlier cleanups (#7633/#7636/#7637 relocations, #7635 broadcast move), the only remaining reverse dependency in non-test source_cell was:

source/source_cell/read_atom_species.cpp
    #include "source_hamilt/module_xc/exx_info.h"   // writes GlobalC::exx_info

read_atom_species parsed the STRU ABFS_ORBITAL / ABFS_JLES_ORBITAL sections and pushed the filenames straight into GlobalC::exx_info.info_ri / info_opt_abfs — the STRU parser (geometry layer) writing into the XC module's global state.

Change

Route the lists through UnitCell, where STRU orbital-file names already live (orbital_fn, descriptor_file):

• UnitCell gains neutral members abfs_orbital_files / jle_orbital_files.
• read_atom_species parses the sections into those members; the exx_info.h include and the cal_exx || rpa guard are removed (an absent section is a no-op via SCAN_LINE_BEGIN).
• bcast_cell broadcasts the two members as part of bcast_unitcell.
• Exx_NAO::init(ucell) copies them into GlobalC::exx_info before Exx_LRI copies info_ri, keeping the EXX-specific routing (which list feeds info_ri vs info_opt_abfs) in the LCAO EXX layer.

Data flow:

read_atom_species → UnitCell members → bcast_unitcell (all ranks) → Exx_NAO::init copies → exx_info → Exx_LRI / opt-ABFs / RPA

Because the lists now travel inside the already-broadcast UnitCell, the dedicated bcast_exx_file_lists() added in #7635 is redundant and removed.

Result

source_cell (non-test) now #includes only source_base and itself — the reverse-dependency goal of #7598 is met. (Test files still cross layers to build fixtures, which is conventional.)

Timing / correctness

init(ucell) runs in ESolver_KS_LCAO::before_all_runners (reused by TDDFT / DM2rho / DoubleXC; LR reuses the moved Exx_LRI), on all ranks, after driver_run has called setup_cell (which reads STRU on rank 0 and broadcasts UnitCell), and before every consumer:

• RI-EXX: the copy is immediately before Exx_LRI copies info_ri.
• opt-ABFs: generate_matrix runs later in the same before_all_runners.
• RPA: RPA_LRI(GlobalC::exx_info.info_ri) is built during SCF, later still.

Behaviour is preserved. The cal_exx guard removal is a no-op in practice: when EXX/RPA is off the ABFS lists are simply unused (Exx_LRI is constructed but not run).

Verification

• Repo-wide grep: non-test source_cell has no remaining exx_info/GlobalC::exx reference; no dangling bcast_exx_file_lists.
• g++ -std=c++14 -fsyntax-only passes on read_atom_species.cpp (with __LCAO), bcast_cell.cpp (with __MPI, stubbed mpi.h), and the init() copy logic compiled against the real exx_info + UnitCell structs. setup_exx.cpp/esolver pull LibRI and are exercised by CI.
• No unit test asserts on the ABFS lists and no STRU fixture contains the sections, so no test behaviour changes.

Minor note

The now-unused rpa parameter of read_atom_species is left in place to avoid rippling the setup_cell signature; it can be dropped in a follow-up.

[Critsium-xy]

Source_cell should be a clean module after this PR.

[mohanchen]

Nice job