refactor(driver): trim compute capability response

[elezar]

Summary

Remove unused GPU availability fields from the internal compute driver capability response. GPU admission remains driver-local and is still validated when a sandbox create request asks for GPU resources.

This intentionally breaks wire compatibility for the internal compute-driver capability message by deleting the removed fields without reserving their field numbers or names.

Related Issue

None.

Changes

• Removed the compute-driver capability proto fields for GPU support and GPU count without reserving their tags or names.
• Stopped populating GPU capability fields from Docker, Kubernetes, Podman, VM, and test drivers.
• Documented that the capability RPC reports driver identity, version, and default image only.
• Addressed clippy fallout from the now-sync Kubernetes capability path.

Testing

• cargo check -p openshell-core -p openshell-driver-docker -p openshell-driver-kubernetes -p openshell-driver-podman -p openshell-driver-vm -p openshell-server
• cargo test -p openshell-server -p openshell-driver-docker -p openshell-driver-kubernetes -p openshell-driver-podman -p openshell-driver-vm --no-run
• mise run pre-commit passes
• Unit tests not applicable; no behavior change beyond removing unused capability fields
• E2E tests not applicable; no sandbox runtime behavior change

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)