PE - FEATURE - Port Fortran sing_spot / use_res_spot for the resonant jump radius (and try layer-width spots for kinetic cases)

[logan-nc]

Problem

The PE singular-coupling jump is evaluated at ψ_s ± spot_psi with a hard-coded numerator (SingularCoupling.jl:378):

spot_psi = 5e-4 / (abs(nn) * abs(sing_surf.q1))

annotated "spot = 5e-4 matches Fortran default singfac_min". That annotation is wrong on both counts: the Fortran quantity is sing_spot, its own namelist knob (default 5e-4, gpec.f:123), unrelated to singfac_min. So the Julia port reproduces the Fortran default but drops the knob — and drops a second capability entirely.

What Fortran actually does

gpec.f:486-508 builds a per-surface sing_spots(ising) array:

• default: sing_spots(ising) = sing_spot (5e-4);
• with use_res_spot = .TRUE. (requires gal_flag): sing_spots(ising) = sing_resspot·(n·|q′|)·sfac^(−1/3) + sing_spot, i.e. the matching radius is set by the resistive resonant-layer width (the S^(−1/3) scaling from singtype(ising)%restype), floored at sing_spot.

Consumed as lpsi/rpsi = respsi ∓ spots(ising)/(nn·|q1|) (gpout.f:598-603, and the same in gpout_singfld:1658-1663). The netCDF output even ships them as sing_spots ("Sweet spots used for calculating resonant jump quantities").

Proposed work

1. Baseline fidelity: expose sing_spot as a control ([PerturbedEquilibrium]), replace the literal, and fix the misleading annotation. Default 5e-4 keeps current behaviour bit-identical.
2. Per-surface resistive spot: port the use_res_spot branch — matching radius from the resistive layer width per surface, floored at sing_spot. The ingredients exist on the Julia side (ForceFreeStates/Resist.jl, the Galerkin Δ′ path, the inner-layer solvers).
3. New work — kinetic cases: try a resonant-layer-width-based spot for kinetic runs, where the relevant layer is not the resistive one. Motivation from issue Performance: Why do large equilibrium splines slow down the code? #376 (handoff/issue376/RESULTS.md §36): on a DIII-D kinetic-calculated case, the resonant-layer surface term is ~20% of the total NTV torque and the PE-vs-NTV comparison agrees to 0.8% only once that layer window is excluded — so the choice of matching radius is not cosmetic there. Related: Feature Request: Integrate self consistent kinetic workflow with the riccati integrator workflow #375 (Riccati + kinetic) and Port gpout_dw / gpout_dw_matrix (psi-resolved torque response profiles and matrices) as the grid-logic diagnostic #423 (torque cross-check).

Also worth noting: a matching-radius bias of exactly this kind was already found and fixed on the Riccati side (#412), which is precedent that the radius choice materially moves Δ′-family numbers.