fix(palace): restore airbox in notebooks so absorbing BC is not on metal

[vvahidd]

Why

palace_microstrip was producing S-parameters far too lossy and reflective for a simple microstrip — S21 = -32.6 dB and S11 = -7.8 dB at 100 GHz, so |S11|² + |S21|² = 0.17 and 83% of the power was being absorbed. This was initially attributed to a Palace 0.17.0 upgrade, but the mesh is built client-side, so a solver upgrade can't remove a material region — and a missing material region is what it turned out to be.

Cause

get_stack(air_above=...) has been deprecated and silently ignored since the airbox refactor (f17f9ca); it only emits a warning, which is easy to miss in a notebook. Several notebooks still passed it and never called set_airbox(), and MeshConfig.airbox_margin defaults to 0.0, so no air region was created at all.

The mesh box then collapses to the design footprint and Palace's 2nd-order absorbing boundary is applied directly to the outer dielectric faces (sio2__None / sin__None) rather than to an airbox. For the microstrip that put the absorbing wall 1.9 µm above the trace (TopMetal2 top z=14.03, box top z=15.93) and flush against the ground-plane edge at y=±27 µm. In the quasi-TEM near field that behaves as a resistive sheet. The missing air also raises ε_eff, pulling Z0 below the 50 Ω ports.

Rebuilding the notebook reproduces the bad run's physical groups exactly (same 11 groups, same numbering); adding set_airbox() restores the air volume and moves the absorbing BC onto air__None.

Changes

• Add set_airbox(margin_x=50, margin_y=50, z_above=100, z_below=100) — matching the values already used by the working CPW/inductor notebooks — and drop the dead air_above from palace_microstrip, palace_branch_line_coupler, palace_cpw_via, palace_width_sweep, and _palace_cpw. palace_cpw_lumped already called set_airbox(), so only its dead argument is removed.
• mesh() now logs a warning when an absorbing boundary exists but no volume is air-like, so this fails loudly instead of yielding plausible-looking but badly wrong S-parameters.

Decisions

• The guard warns from mesh() only — not write_config() / validate_mesh() / run() — so it fires once, before any cloud submit.
• Used logger.warning rather than warnings.warn to match the local idiom in mesh() (the adjacent >75k-node advisory and the existing set_stack(air_above=...) deprecation).
• An earlier version recorded outer-face names in groups.py to check which material each absorbing face touched. Simplified to checking groups["volumes"] for an air-like region, which needs no new plumbing. Slightly weaker: a stack with an internal air gap but no airbox would not be flagged.
• Un-deprecating get_stack(air_above=...) was considered and rejected in favour of migrating callers to the current API.

Note

Notebook outputs are not regenerated and still show the old values; the affected notebooks need a re-run to refresh them.