feat(cctv): seven camera packs, semantic labels, terrain-clear monitor planes (#433)

Seven contributed CCTV camera packs (Ontario 511, Fintraffic Finland, DriveBC, TxDOT Texas, Estonia Tallinn + Tarktee, Warendorf, Live Traffic NSW) on a fair per-pack catalog cap; semantic camera labels; monitor planes lifted clear of the terrain from a precomputed, bundled ground-height dataset. Default catalog 800 → 3,446 cameras. Full context in PR #433.

Co-authored-by: sendmebits <[email protected]>
Co-authored-by: avocadoAgent <[email protected]>
Co-authored-by: Alex Harvey <[email protected]>
Co-authored-by: Jason Meredith <[email protected]>
Co-authored-by: Daniel Smith <[email protected]>
Co-authored-by: Vladimir Sobolev <[email protected]>
Co-authored-by: Matthias Schneider <[email protected]>
Co-authored-by: Cyber Knight <[email protected]>
Co-authored-by: NovaSoft Australia <[email protected]>

src/data/local_data/cctv_ground_heights/README.md

@@ -0,0 +1,58 @@
+# CCTV ground heights
+
+Precomputed ground height under every camera in the served CCTV catalog and
+under the nine support points of its monitor plane, so the client can place a
+camera and its plane with no runtime sampling. Shared placement data: any
+consumer that renders these cameras on the same photorealistic surface can
+reuse it.
+
+Source surface: Google Photorealistic 3D Tiles, sampled at maximum detail with
+CesiumJS `scene.sampleHeightMostDetailed` from a running God's Eye View
+(scripts/precompute-cctv-heights.mjs). Heights are WGS84 **ellipsoidal**
+metres; no geoid conversion is needed or applied. This is placement data
+derived from the rendered surface, not imagery.
+
+Runtime output: `cctv_ground_heights.json`
+
+```
+{
+  "schemaVersion": 1,
+  "provider": "google-photorealistic-3d",
+  "heightReference": "WGS84-ellipsoid",
+  "generatedAt": "<ISO>",
+  "cameras": {
+    "<camera id>": {
+      "poseHash": "p1-…",          // hash of the served pose the samples describe
+      "status": "ok" | "miss",
+      "mountGroundM": <number>,    // ground under the camera position (absent on a miss)
+      "supports": {                // ground under the plane's 3×3 support grid; a value is
+                                   // null where that point had no sample (absent on a miss)
+        "bl","bm","br",            // bottom row (left, middle, right)
+        "ml","mc","mr",            // middle row
+        "tl","tm","tr"             // top row
+      },
+      "misses": [<keys with no sample: "mount" and/or support keys>],
+      "sampledAt": "<ISO>",
+      "attempts": <n>
+    }
+  }
+}
+```
+
+Support point positions are defined by `src/data/cctvFootprint.js`
+(`planeSupportPoints`), which the client, the server join
+(`server/providers/cctv/groundHeights.js`) and the precompute all share. An
+entry is used only while the camera's served pose still hashes to
+`poseHash`; a camera whose feed moved it, or whose pose a user edited, falls
+back to runtime placement from the Re:Earth DEM.
+
+Regenerate (resumable; only cameras whose pose changed or whose entry is
+incomplete are re-sampled):
+
+```
+GEV_BASE=http://localhost:4173 node scripts/precompute-cctv-heights.mjs
+```
+
+Coverage at generation (2026-09-14): 3,445 of 3,446 cameras `ok` (six of them
+with one or more null supports), 34,444 sampled heights, 43 minutes on a real
+GPU. Regenerate whenever a pack changes its poses or a new pack lands.