feat(nvml-mock): vGPU host/guest simulation

[giuliocalzo]

Priority: Low

Context. vGPU is used in virtualized GPU deployments.

Gap. No vGPU host or guest mode in the NVML mock (nvmlVgpu* are stubs).

Proposed scope.

• Implement core nvmlVgpu* / nvmlDeviceGetVgpu* calls backed by config.
• E2E: a vGPU-aware consumer enumerates configured vGPU instances.

Why. Extends NVML coverage (relates to #254) toward virtualization scenarios. Low priority.