Cleanup and refactor materials library

[oskooi]

Context

python/materials.py (1682 lines) is Meep's optical materials library — a flat, import-time-evaluated module that defines ~60 mp.Medium objects (semiconductors, dielectrics, metals, birefringent crystals). It is a public API: users do from meep.materials import Si, Au, SiO2, ... and the names appear in tests, examples, tutorials, and doc/docs/Materials.md.

The file works but has accumulated problems that make it error-prone to extend:

• Dead code: import numpy is never used.
• Typo: line 813 comment reads "chroimium" (should be "chromium").
• Order-dependent shadowing bug (latent): metal_range is reassigned 3× at module scope (lines 294, 338, 380). The 8 Rakic metals defined after Cu — Al, Be, Cr, Ni, Pd, Pt, Ti, W — silently inherit Cu's range because that was the last assignment. Reordering or inserting a metal would silently change valid ranges.
• Intra-material variable shadowing: birefringent crystals reuse the same globals for the ordinary and extraordinary axes (e.g. BaB2O4_frq1, LiNbO3_frq1, Al2O3_frq1, SiO2_frq1, etc., are each assigned twice). Correct only because susceptibility lists are built eagerly between the two blocks — fragile and confusing.
• Namespace pollution / no __all__: hundreds of intermediate globals (X_frq1, X_gam1, X_sig1, X_susc, X_range) leak into meep.materials's namespace. import * exports all of them.
• Massive boilerplate: each material repeats the same ~10-25 lines. Metals additionally repeat the Rakic formula sigN = fN * plasma**2 / frqN**2 ~5× each.
• No programmatic lookup: materials can only be selected by hand-typed name; there is no name→Medium registry.

Goal: refactor for readability and safety while keeping the public API and the exact numerical output identical, and add an additive registry for name-based lookup.

Hard constraints (must not change)

• Every existing material name stays (E.g. Si, Au_JC_visible, SiO2_aniso, etc.)
• Public constants um_scale and eV_um_scale stay at module top, before any material is built (docs instruct users to edit um_scale in place — this must keep working since materials are built at import using its value.)
• Every mp.Medium must produce numerically identical epsilon(freq) to the current file (target: exact; tolerance ≤ 1e-12).
• Preserve all reference comments (papers, refractiveindex.info URLs, wavelength ranges), and ## WARNING: unstable; field divergence may occur notes (Ag_visible, Al_visible, Mo).