Expose enable-cuda-compat as an API under `nvidia-container-runtime`

**What I'd like:**
I would like to expose an API that allows users to opt-in or opt-out from the `enable-cuda-compat` feature provided by the NVIDIA Container Toolkit and the Kubernetes Device Plugin, given the [recent security findings] around the feature.

The API should be used both in the NVIDIA Container Runtime configuration file, and in the NVIDIA k8s Device Plugin, to disable using the hook while generating the CDI specifications.

[recent security findings]: https://www.wiz.io/blog/nvidia-ai-vulnerability-cve-2025-23266-nvidiascape