source_basis/module_ao: orb_atomic_lm_test off #define private public, via existing accessors plus one friend

[Critsium-xy]

Continues the #define private public cleanup (#7940, #7949, #7952). This one
takes orb_atomic_lm_test.cpp, the largest single concentration of the pattern
left in the tree.

What the macro was covering

The file reaches into 17 private members of Numerical_Orbital_Lm and calls
four of its private methods. No PARAM is involved. Sixteen of the seventeen
members already have public accessors, so most of the macro's job was to let
the test bypass an interface that was already there.

138 sites now read through those accessors — get_psi(), getNr(),
getRcut(), getL() and the rest. The substitution is uniform because each
accessor returns the member itself: get_psi() yields
const std::vector<double>&, so .psi[i], .psi.size(), .psi.empty() and a
bare .psi all keep working through it, and the scalar getters return const
references. Nothing about what the test asserts changes.

Two things needed more than a substitution, and they are deliberately handled
differently.

psir — completing an accessor set, not inventing one

Numerical_Orbital_Lm exposes a systematic trio for each array:

const double*  getPsi() const                 { ... }   // pointer
const double&  getPsi(const int ir) const     { ... }   // element
const std::vector<double>& get_psi() const    { ... }   // vector

psi, psif, psik, psik2, r_radial and rab all have all three. psir
had only the pointer and element forms, so .psir.size() and .psir.empty()
had no public route. This adds the missing get_psir(), which fills an obvious
gap in the class's own pattern rather than adding an accessor because a test
wanted one.

The four private methods — a named friend grant

cal_kradial, cal_kradial_sbpool, cal_rradial_sbpool and plot are private,
and exercising them is precisely why four of these tests exist (the file header
lists them under "Tested functions"). They get
friend class NumericalOrbitalLmTest;, placed next to the friend class Numerical_Orbital; the class already carries.

Friendship is not inherited and a TEST_F body lives in a class derived from
the fixture, so the fixture gained four forwarding wrappers and the bodies call
those — the arrangement already used by dftu_lcao_test.cpp and introduced for
this cleanup in #7949.

Why not simply grant friendship for everything

A friend would have made all 138 sites compile untouched, and the diff would
have been a dozen lines. It would also have left the test reaching past an
interface that already exposes exactly what it needs. The narrower reading is
that friendship is for what genuinely has no public route — here, four private
methods — and the accessors are for everything else.

Result

	before	after
files with the macro (tree-wide)	63	62
occurrences	91	90
private members of Numerical_Orbital_Lm named in a TEST_F body	17	0

Production change is 4 lines: one accessor and one friend declaration.

Verification

Linux, cmake -B build -G Ninja -DBUILD_TESTING=ON -DENABLE_LCAO=ON -DENABLE_MPI=ON -DENABLE_OPENMP=ON,
then cmake --install build — the module_ao tests take their orbital data
from install(DIRECTORY lcao_H2O ...), so without that step ORB_read_test
fails and ORB_atomic_lm_test / ORB_nonlocal_lm_test segfault on missing
input, both before and after this change.

• build: 0 errors.
• MODULE_AO_ORB_atomic_lm_test and the three sibling module_ao tests: all pass.
• full unit suite: 2 of 339 fail, the same two as upstream/develop
  (MODULE_HSOLVER_diago_hs_parallel, MODULE_HSOLVER_LCAO, both
  mpirun-based). No new failures.
• agent_governance_check.py --base upstream/develop --head HEAD: 0 errors.
  The access-hack ratchet reports nothing (1 removed, 0 added), and no
  PARAM/GlobalV/GlobalC reference is added or removed.
• After the rewrite, checked mechanically that no private member of
  Numerical_Orbital_Lm is named anywhere in a TEST_F body, and that the file
  contains no std::swap, address-of or assignment form that would have slipped
  past a read-only substitution.

No INPUT parameter and no user-visible behaviour changed, so
docs/parameters.yaml and docs/advanced/input_files/input-main.md need no
update.

What is not here

orb_nonlocal_lm_test.cpp is the last module_ao file carrying the macro. It
needs a different treatment and is left for a separate PR: besides the read-only
substitutions, it deliberately mutates internals — swapping the r-space and
k-space arrays of a projector, reallocating rab, then calling the private
get_kradial() to check round-trip consistency — and asserts on raw pointers
being nulled by freemem() / restored by renew(). That needs mutating
friend wrappers, which is a different review question from the read-only
forwarding here, and is better expressed as one wrapper for the whole
r-to-k swap than as per-field access.

🤖 Generated with Claude Code

[mohanchen]

LGTM