bug: GPU vendor displays "nvidia" when accelerator is metal

[Defilan]

When deploying with --accelerator metal, the deploy output shows GPU: 1 x nvidia (layers: 26) even though the deployment is targeting Apple Silicon with Metal.

The --gpu-vendor flag defaults to nvidia and the deploy summary doesn't override the display when Metal is selected.

Steps to reproduce:

llmkube deploy llama-3.2-3b --gpu --accelerator metal

Expected: GPU: 1 x apple (layers: 26) or omit the vendor line entirely for Metal deployments.

Actual: GPU: 1 x nvidia (layers: 26)

Workaround: Pass --gpu-vendor apple explicitly.

This is cosmetic only. The deployment works correctly with Metal regardless of the displayed vendor.