fix(gpu): select single CDI GPU defaults

[elezar]

Summary

Builds on #1156 to implement Docker and Podman CDI GPU selection for resource_requirements.gpu.

Default GPU requests select one concrete NVIDIA CDI device. Count requests select count concrete devices. Explicit device_ids pass through unchanged.

Related Issue

Closes #1477

Blocked by #1156.

Changes

• Adds shared CDI inventory and round-robin selection with count-aware checks.
• Docker uses daemon CDI inventory from DiscoveredDevices.
• Podman maps local /dev/nvidiaN devices to CDI IDs.
• Updates GPU docs and optional GPU e2e expectations.

Testing

• cargo check -p openshell-core -p openshell-cli -p openshell-server -p openshell-driver-kubernetes -p openshell-driver-docker -p openshell-driver-podman -p openshell-driver-vm --tests
• cargo test -p openshell-core -p openshell-cli -p openshell-server -p openshell-driver-kubernetes -p openshell-driver-docker -p openshell-driver-podman -p openshell-driver-vm gpu --tests
• mise run pre-commit

Checklist

• Follows Conventional Commits
• Documentation updated

[github-actions]

🌿 Preview your docs: https://nvidia-preview-pr-1675.docs.buildwithfern.com/openshell

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.