fix(#278) warn when fixed_psi_* target is unreachable; add reference.natural_psi

[prjemian]

• closes psic fixed_psi_horizontal: NoForwardSolutions for every hkl on hexagonal lattice (sibling of #275) #278

Investigation

#278's "0 solutions" symptom is mathematically correct behavior under the \u03c8-validation-filter design (issue #176): for fixed (UB, hkl), every Bragg solution has the same natural \u03c8 \u2014 uniquely determined by Q_phi = UB @ hkl and the azimuthal reference. The user asked for psi=0 on reflections whose natural \u03c8 are 90\u00b0, 107\u00b0, 180\u00b0, etc., so forward() correctly returned []. The user's own Newton check in the issue confirmed these reflections aren't reachable at psi=0. The real defect was the silent empty list.

Changes

• _solve_psi_mode now emits a UserWarning that names the natural \u03c8 and points the user at ad_hoc_diffractometer.reference.natural_psi when the target mismatches; a separate warning fires when \u03c8 is undefined (Q parallel to reference or beam).
• New public helper reference.natural_psi(geometry, h, k, l) returns the natural \u03c8 from UB and hkl alone (None when undefined).
• YAML comments above every fixed_psi* mode in psic, kappa6c, fourcv, fourch, kappa4cv, and kappa4ch clarify the validation-filter semantics.

Recommended workflow (now possible)

nat = ahd.reference.natural_psi(g, h, k, l)
if nat is None:
    # \u03c8 undefined for this reflection \u2014 pick another or change azimuthal_reference
    ...
for c in g.modes['fixed_psi_horizontal'].constraints:
    if isinstance(c, ahd.ReferenceConstraint) and c.name == 'psi':
        c._value = nat
sols = g.forward(h, k, l)   # now returns real solutions, no warning

QC

• 2472 \u2192 2493 tests pass; 100% coverage maintained
• pytest -m slow_benchmark: 3 passed
• pre-commit, ruff, US English check all clean

Contributed by: OpenCode (argo/claudeopus47)