example(RTVolume): GPU-written device-buffer procedural AABBs (#37)

A compute shader (aabbs.comp.{glsl,wgsl}) writes the procedural box into a
device buffer that BuildProcedural consumes by device address/handle with no
host copy; the WebGPU path also refits from it each frame so the box breathes.
Verified rendering on both Vulkan (native) and WebGPU (browser).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

examples/RTVolume/aabbs.comp.glsl

@@ -0,0 +1,40 @@
+#version 460
+#extension GL_EXT_buffer_reference : enable
+#extension GL_EXT_scalar_block_layout : enable
+
+// Issue #37: GPU-resident procedural AABBs. This compute shader is the
+// "producer" — it writes the per-box bounding boxes straight into a device
+// buffer that Mesh::BuildProcedural / RefitProcedural then consume by device
+// address, with no host copy. Here it emits a single unit box [-1,1]^3 (the
+// bounding volume the intersection shader turns into a sphere); `time` lets
+// the box breathe so a per-frame refit has something to track. A real
+// particle system would write one box per particle from simulation state.
+//
+// The buffer is reached via a buffer_reference (its device address pushed in
+// `pc`), so this dispatch needs no descriptor-heap binding for the output.
+
+layout(buffer_reference, scalar) buffer AabbRef {
+    float data[];   // 6 floats per box: min.xyz, max.xyz (RTAabb layout)
+};
+
+layout(push_constant) uniform PC {
+    AabbRef boxes;
+    uint    count;
+    float   time;
+} pc;
+
+layout(local_size_x = 64) in;
+
+void main() {
+    uint i = gl_GlobalInvocationID.x;
+    if (i >= pc.count) return;
+
+    float r = 1.0 + 0.15 * sin(pc.time);   // gently pulsing radius
+    uint b = i * 6u;
+    pc.boxes.data[b + 0u] = -r;
+    pc.boxes.data[b + 1u] = -r;
+    pc.boxes.data[b + 2u] = -r;
+    pc.boxes.data[b + 3u] =  r;
+    pc.boxes.data[b + 4u] =  r;
+    pc.boxes.data[b + 5u] =  r;
+}