feat(web-standalone): antenna height from the tallest building within 50 m of a repeater

[khagele]

Closes #74, the antenna-height half. The other half, building geometry in the line-of-sight profile, is measured and not built; numbers in #74 (comment): dense sampling calls 6 of 10 real receptions blocked or partial, so more resolution penalises the true cell under the current knife-edge model.

The line-of-sight check put every observer's antenna at 12 m. Sampled against 3D BAG, 32 of 60 fixture observers stand within 12 m of a building, roof median 10.2 m above ground, p90 25 m; the constant is right for about half of them and low for a quarter.

Change

• bag3d.py (stdlib, tested): WGS84 to RD New and back (within 1 m on the Westertoren and the Martinitoren), the buildings in a 3D BAG CityJSON response with their footprints in metres via the response transform, distance from a point to a footprint (0 inside), and the tallest building whose footprint comes within a radius, ties to the nearer one.
• server.py: /proxy/3dbag/roof?lat=&lon=, a 100 m box around the node, up to 4 pages of 100 features, radius 50 m, cached for a day per position; no roof outside the Netherlands; 400 on bad input.
• On activation the page looks up the same nodes the link budget covers and sets roofHeightM and roofDistanceM; terrainLosFactor() starts the beam at roof plus a 3 m mast for those and at 12 m for the rest. The cluster status says how many; the observer popup shows the antenna height, its source and the distance to that roof; a dragged node drops its roof with its radius.
• Dockerfile and .dockerignore ship the module; README and AGENTS name the source; changelog entry.

Why 50 m and the tallest, not the footprint underfoot: advertised positions are metres off and a mast can stand next to the building it serves. At 100 m, 3 of 8 observers on the test region picked a taller building 60 to 90 m away; at 50 m they keep the one beside them.

Verified

• pytest web-standalone/tests: 13 passed.
• Live route: 48.7 m in the Amsterdam centre (2 pages, 7.5 s first time, 0.01 s cached), 75.9 m on the Westertoren, 11.7 m at 79 m for a suburban node at 100 m radius; no roof for Paris; 400 for lat=x.
• In the page, region 1 of the fixture deep link: 14 lookups, 13 of 14 observers have a building within 50 m, the farthest at 47.2 m; antennas from 10.6 m (a 7.6 m roof 23 m away) to 39 m (a 36 m roof underfoot); status "13 of 14 observers have a building within 50 m per 3D BAG and use its roof, the rest assume 12 m"; no console errors.

Not measurable

The effect on estimates: the accuracy harness does not run the terrain pass (#68, #72). This lands on the argument that a 25 m antenna at 25 m beats a 25 m antenna at 12 m in a diffraction profile, not on a number.

🤖 Generated with Claude Code