Add electrostatic no-op benchmark case with unconverged-reference finding

[krystophny]

Benchmark slice of #175 (no-op comparison against the hat basis), with the
honest outcome: the benchmark cannot gate anything yet, and the committed
case documents why.

KIM/tests/test_data/benchmark_electrostatic/ contains a self-contained
electrostatic case (Krook, the profile set of the golden electromagnetic
case), a periodic-solver twin configuration whose layer covers the widest
admissible window for these profiles, compare_phi.py (raw and
gauge-offset-removed L2/Linf on a common window, threshold 5 percent, exit
nonzero on failure), the two windowed profiles, and a README with the
measured numbers and the manual two-command procedure.

Stack: parent feat/flux-pumping/wp5-driver at fb07a293 (#180).

Finding

Both solves are unconverged on this case, so no agreement threshold is
meaningful:

• hat 128-grid vs periodic 128-mode on the flat layer: max |Phi| differs
  by ~11 orders (1.741e10 vs 2.132e-1 statV); offset-removed relative L2
  0.975; an optimal complex rescaling still leaves 0.976, so the shapes
  are uncorrelated, not misnormalized.
• Mode doubling 128 -> 256 changes the periodic solution by relative L2
  1.072 (not converged).
• Grid doubling 128 -> 256 changes the hat reference by relative L2 1.0
  with the peak growing by 3.6e3: the Krook hat solve diverges under
  refinement at omega = 0, consistent with the configuration file's own
  "Krook (unstable)" annotation. There is currently no refinement-stable
  reference.

No ctest is wired: a threshold on two unconverged solutions either always
fails or encodes the defect. The comparison script and case stay as the
review surface and as the ready harness for the follow-up.

Follow-up this defines

Either a Krook case in a parameter regime where the hat solve is
refinement-stable (finite frequency or higher collisionality), or
continuum FokkerPlanck-form kernels for the periodic side, before the
no-op condition of #175 can be tested. Solver code is deliberately
untouched here.

Verification

$ ctest --test-dir build
100% tests passed, 0 tests failed out of 30
$ python3 compare_phi.py <hat>/fields/Phi.dat <periodic>/fields/Phi_periodic.dat
exit 1, relative L2 0.975 (threshold 0.05)

Relation to Markus's Fokker-Planck periodic branch

This PR is retained as a negative benchmark record and reusable common-grid comparison harness. It proves that the packaged Krook hat solution is not a converged no-op reference; it does not validate either periodic implementation. Do not tune this case further. The accepted matched FP/Vaclavik benchmark is #190.