v0.8.0: map roads/sidewalks, per-vertex depth bias, retail camera controls

- ra3.map: capture the object RoadType; ra3.models pairs the map's
  Start/End road objects and drapes a flat textured ribbon along each
  segment (tessellated to the relief).
- Ground decals (roads/sidewalks) get a per-vertex screen-space depth
  offset (scene::vertex.bias) applied in the object shader and the
  software rasteriser, so they no longer z-fight or clip into the
  terrain at top-down angles.
- Tactical view aligned with retail (SAGE LookAtTranslator): middle-drag
  orbits yaw/pitch, a middle click resets angle/pitch/zoom, arrow keys
  pan, wheel zooms within the retail zoom range; WASD pan removed and
  left-drag no longer rotates.
- ui_event gains middle/right/released buttons; Vulkan + SDL map them.
- docs: roads, depth bias and camera controls.

src/terrain/ra3.terrain.cppm

@@ -841,9 +841,11 @@ export namespace ra3::terrain {
                         const auto inv_a = w0 * p0.inv_a + w1 * p1.inv_a + w2 * p2.inv_a;
                         const auto depth = 1.0F / inv_a;
                         const auto pixel = static_cast<usize>(y) * static_cast<usize>(rw) + static_cast<usize>(x);
-                        // Ground decals sit exactly on the terrain; a small bias
-                        // (mirrors the GPU depth bias) keeps them from z-fighting.
-                        if (depth - 0.5F >= zbuf[pixel]) continue;
+                        // Ground decals carry a per-vertex bias toward the camera
+                        // (the retail shader's screen-space depth offset); it
+                        // keeps roads/sidewalks from z-fighting the terrain.
+                        const auto bias = w0 * v0.bias + w1 * v1.bias + w2 * v2.bias;
+                        if (depth - bias - 1.0F >= zbuf[pixel]) continue;
 
                         const auto tu = (w0 * v0.u * p0.inv_a + w1 * v1.u * p1.inv_a + w2 * v2.u * p2.inv_a) * depth;
                         const auto tv = (w0 * v0.v * p0.inv_a + w1 * v1.v * p1.inv_a + w2 * v2.v * p2.inv_a) * depth;