Add periodic Fourier kernel assembly and solve

[krystophny]

Third split item of #175: the discrete solve on the periodized modes.
fourier_periodic_m (KIM/src/fourier_periodic/fourier_periodic_m.f90)
provides the mode grid k_m = 2 pi m / L, the matrix elements
K_{m,m'} = (2 pi / L) times the one-period equidistant r_g quadrature of
the restored kernels, the dense LAPACK solve of the screened Poisson system
(k_m^2 delta - 4 pi K^{rho Phi}) Phi = rhs, and the inverse expansion
f(r) = sum_m f_m exp(i k_m r).

Stack: parent branch feat/flux-pumping/wp5-periodization at 4671fb10
(#178), on top of #177; the diff is cumulative until the parents merge.
Driver integration behind a config switch and the hat-basis/Vaclavik
benchmarks are the remaining split items; nothing calls this module yet.

Invariant preserved

• Absolute-coordinate phases: the assembly quadrature and the inverse
  expansion use the same exp(i k r) convention as the
  exp(i (k_r - k_r') r_g) factor inside the kernels, so the uniform
  background is exactly diagonal through discrete Fourier orthogonality
  of the equidistant midpoint rule (exact for |m - m'| < N_quad).
• The uniform diagonal reproduces the adiabatic screening value
  -sum_s lambda_{D,s}^{-2} / (4 pi), so the system matrix carries
  k_m^2 + sum_s lambda_{D,s}^{-2}: the sign convention of the historical
  k-space Debye demonstrator. The kinetic term regularizes the k_0 = 0
  mode, so no separate gauge handling is needed.
• Midpoint quadrature is the lowest-order rule the note prescribes for the
  periodic integrand; no interpolation, truncation, or padding enters.
• CGS-Gaussian units; existing solver paths and golden records untouched.

Verification

Test fails on the flipped kinetic sign

$ sed -i 's|system = -4.0d0*pi*kernel_rho_phi_matrix|system = 4.0d0*pi*kernel_rho_phi_matrix|' fourier_periodic_m.f90
$ ctest --test-dir build -R test_fourier_periodic --output-on-failure
1/1 Test #26: test_fourier_periodic ............***Failed    0.08 sec
 FAIL mode solve matches the screened solution

Test passes after restore

$ ctest --test-dir build -R test_fourier_periodic
1/1 Test #26: test_fourier_periodic ............   Passed    0.06 sec

Full suite

$ ctest --test-dir build
100% tests passed, 0 tests failed out of 29

Checks (1e-12 relative): centered mode grid, exact uniform diagonality,
adiabatic screening diagonal on every mode, the screened point-charge
solve against the analytic per-mode solution, L-periodicity of the
expansion, and a real potential at the charge location. Quadrature
convergence on a radially varying background is exercised in the driver
integration item, where real profiles enter.

Relation to Markus's Fokker-Planck periodic branch

This PR assembles and solves the restored Krook kernels. Its mode-grid, quadrature, dense-solve, and manufactured-Debye tests may be reusable infrastructure, but its kernel and Poisson conventions are not an authority for the FP branch. The matched FP boundary-value problem is #188; final selection is #191.