Class: WebGpuSimulationRun
WebGpuSimulationRun()
A resumable megakernel WebGPU run. Source generation and membership are one dispatch. Each trace invocation then keeps its current ray in local state for several interactions, appends extra branches to a slot-major output half, and resumes from a stable-compacted queue in the other half.
A WebGPU run is GPU-authoritative. Environments without WebGPU must select CpuSimulationEngine explicitly rather than silently executing another backend through this class.