ForceFreeStates - Investigate Solovev cutoff sensitivity after 2.0 #610

Follow-up to #555, scheduled for investigation after the 2.0 release. This issue records numerical questions to investigate without holding #555 up on additional convergence studies.

Related: #481 (separating crossing width from matching distance), #490 / #520 (forward-basis conditioning), and #427 (PE singular-coupling sampling radius). This issue focuses on validation of the Solovev examples rather than proposing another crossing implementation.

singfac_min controls the ForceFreeStates (FFS) integration/crossing near singular surfaces. Changing it can change the global solution and energy matrix, hence the response computed by PerturbedEquilibrium (PE), as well as downstream NTV torque. It is distinct from the PE singular-coupling diagnostic's sampling offset and from regularization_width.

The ideal-plus-NTV Solovev examples use kinetic_factor = 0 and prescribed regularization. Their torque should be interpreted as a diagnostic on a regularized ideal response, not a self-consistent kinetic torque. Numerical convergence at fixed regularization and physical sensitivity to the regularization are separate questions. The calculated-kinetic path (kinetic_source = "calculated", kinetic_factor = 1) needs separate validation.

After 2.0:

• Revisit the Solovev cutoff scan and its reproducibility, including ODE-tolerance sensitivity and the conditioning concerns tracked in ForceFreeStates - BUGFIX - Decouple the ideal-crossing jump width from singfac_min #481 and ForceFreeStates - BUGFIX! - Trigger the Gaussian reduction on the conditioning of U₁ and factor U₁ in the plasma response matrix #520.
• Test an intermediate-cutoff window around 1e-6 with independent ODE-tolerance and spatial-resolution checks. Track FFS energies, response, and coupling quantities; treat regularized NTV torque as a downstream diagnostic.
• Separate finite-crossing error, small-cutoff numerical breakdown, and prescribed-regularization dependence. State the acceptable error for the quantity being validated.
• Validate the calculated-kinetic examples independently rather than extrapolating from ideal-plus-postprocessing runs.
• Add focused regression coverage and document the supported numerical range when established.

Treat 1e-6 as a practical choice to validate further, without assuming that smaller values are always safer or that 1e-4 is more accurate. No solver design is selected by this issue. This follow-up does not replace the normal human review and merge checks for #555.