feat: Palace config-only features and new sim types

[vvahidd]

Summary

Low-hanging Palace features that are purely config-driven (just adding JSON keys to config.json) plus wiring up existing config models. Based on review of Palace docs.

Config-only features

These only require adding fields to existing models and populating the Palace JSON — no mesh or geometry changes needed.

• Field saving — expose Save/SaveStep on DrivenConfig, EigenmodeConfig for ParaView .vtu export
• Domain Energy postprocessing — add_energy_domain("substrate") → Domains.Postprocessing.Energy (energy per volume per frequency)
• Probe postprocessing — add_probe("name", center=[x,y,z]) → Domains.Postprocessing.Probe (E/B field at a point per frequency)
• Surface Flux postprocessing — power/E/B flux through mesh surfaces → Boundaries.Postprocessing.SurfaceFlux
• London depth — set_material("al", london_depth=...) → Domains.Materials[].LondonDepth for superconductor kinetic inductance
• PMC boundary — add_pmc() → Boundaries.PMC for symmetry planes (halves sim time)
• Adaptive circuit synthesis — circuit_synthesis=True on DrivenConfig → AdaptiveCircuitSynthesis (extract lumped L/R/C from PROM)
• Uniform refinement levels — expose Model.Refinement.UniformLevels on MeshConfig
• Box/Sphere refinement — add_refinement_box() / add_refinement_sphere() for local mesh refinement
• Impedance boundary — add_impedance(Rs=..., Ls=..., Cs=...) → Boundaries.Impedance
• Dielectric interface loss (EPR) — add_dielectric_interface(type="MA", ...) → Boundaries.Postprocessing.Dielectric for quantum loss participation
• Verbose/log level — expose Problem.Verbose (currently hardcoded to 3)

Wire up existing models

Config models already defined in models/problems.py but not yet connected to sim classes.

• Transient sim — TransientConfig exists, needs TransientSim class + cloud run()
• Magnetostatic sim — MagnetostaticConfig exists, needs MagnetostaticSim class + cloud run()

Notebooks

• Adaptive vs uniform sweep comparison — demonstrate speed/accuracy tradeoff

[vvahidd]

Circuit synthesis — not worth implementing

Investigated AdaptiveCircuitSynthesis (extracts L⁻¹, R⁻¹, C matrices from the PROM). After testing, this is not worth wiring up because:

1. Incompatible with Conductivity BCs — Palace requires PEC conductors, no finite-conductivity boundaries, and absorbing BC order ≤ 1. Our default config uses Conductivity boundaries for metal layers + order-2 absorbing, so circuit synthesis fails out of the box.
2. Loses conductor loss — forcing PEC to satisfy the constraint means you lose the resistive loss that matters for RF electrode modeling.
3. S-param fitting is better — taking the S-parameters from a standard driven sim and fitting RLGC via scikit-rf is simpler, includes conductor loss, and plugs directly into existing workflows (e.g., traveling-wave MZM equivalent circuit models).

Recommend removing the "Adaptive circuit synthesis" checkbox from this issue.