blekthsety52-dev/sandbox-device-plugin ? reverse-engineered prompt

Build me a Kubernetes device plugin for NVIDIA GPUs that are set up for passthrough, so I can attach them to Kata VM based containers in a confidential containers cluster.

I want it to find GPUs on a node that are bound to vfio pci, expose them as allocatable devices in Kubernetes, and do a basic health check on them. Please include a simple way to deploy it as a daemonset, plus example YAMLs showing how to run a VM with a GPU attached. If it helps, look up current docs online for Kata and Kubernetes device plugins, since I want this to fit modern setups.

Make sure the project can be built into a binary and container image, and keep the setup instructions clear enough that someone can follow them to prepare the host, enable IOMMU, load vfio pci, and verify the GPU is ready.