feat(eim): variational effective-index method for 2D FDTD #156

[thanojo]

Closes #156 by adding a variational effective-index method (varEIM) that derives the in-plane permittivity for 2D FDTD from the vertical slab mode of the 3D stack (Hammer & Ivanova 2009), instead of bulk indices.

• src/gsim/meep/eim.py: variational_effective_permittivity(), 1D finite-difference slab solver, build_eps_z from the layer stack, and a zero-dispersion fit_medium stub.
• nbs/meep_ybranch_veim.py: Y-branch 2D varEIM vs 3D comparison notebook (+ docs nav entry).

This is a first cut — more to do:

• Dispersive fit: eps_eff(lambda) -> Sellmeier -> Lorentzian poles (replace the fit_medium stub).
• Integrate into the solver workflow (e.g. sim.solver(is_3d=False, eim="variational")).
• Tests against the paper examples (Bragg grating, ring FSR, strip-to-slot counter-example).
• TM polarization (different weight kernel); full spatial eps_eff(x,y) map.

Scope now: TE-only, single reference mode, single wavelength.

Separate finding (not in this PR): 3D waveguide-port S-params under-read by several dB since the Apr-22 "right-size port monitors" change — the in-plane port_margin undersizes the 3D eigenmode source/monitors. Worth its own issue.