feat(cli): add GPU count requests

[elezar]

Summary

Adds CLI and API support for counted GPU sandbox requests. --gpu-device and --gpu-count both imply a GPU request, and Kubernetes receives counted nvidia.com/gpu resource limits while unsupported drivers reject count-based selection explicitly.

Related Issue

Part of #1444. Related to #1338 and #1156. Follow-up GPU support preflight semantics are tracked in #1807.

Changes

• Make --gpu-device imply --gpu in CLI-created sandbox requests.
• Add --gpu-count to the CLI, public sandbox proto, and compute-driver proto.
• Normalize GPU request intent in the gateway when gpu_count or gpu_device is present.
• Render Kubernetes nvidia.com/gpu limits from gpu_count and reject explicit gpu_device requests for Kubernetes.
• Reject unsupported gpu_count requests in Docker, Podman, and VM driver paths, while allowing VM gpu_count = 1.
• Update GPU request docs and driver architecture notes.

Testing

• mise run pre-commit passes
• Unit tests added/updated
• E2E tests added/updated (if applicable)

Focused checks also run during local review:

• mise x -- cargo test -p openshell-cli gpu_count
• mise x -- cargo test -p openshell-cli gpu_device
• mise x -- cargo test -p openshell-server gpu_count
• mise x -- cargo test -p openshell-server gpu_device
• mise x -- cargo test -p openshell-driver-kubernetes gpu
• mise x -- cargo test -p openshell-driver-docker gpu_count
• mise x -- cargo test -p openshell-driver-podman gpu_count
• mise x -- cargo test -p openshell-driver-vm gpu_count

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)
• Architecture docs updated (if applicable)

[github-actions]

🌿 Preview your docs: https://nvidia-preview-pr-1812.docs.buildwithfern.com/openshell

[mrunalp]

/ok to test abe5b79

[TaylorMutch]

/ok to test abe5b79