Show host GPU driver on fleet instances

[peterschmidt85]

Problem

When dstack provisions instances, the accelerator driver installed on the host (NVIDIA/AMD/Tenstorrent) is not visible anywhere — anyone listing fleet instances cannot tell which driver a host runs.

Solution

Detect the driver on the host and show it on fleet instances. When it cannot be detected, the field stays None.

• dstack-shim detects the driver version once at startup, inside the existing GPU detection: an extra driver_version column in the nvidia-smi query, --driver added to the existing amd-smi static call (with a fallback retry for older amd-smi), and /sys/module/tenstorrent/version for Tenstorrent.
• The shim healthcheck response gains optional gpu_vendor/gpu_driver_version fields. The endpoint, its polling, and the handler cost are unchanged — the handler reads only the GPU list cached at startup.
• The server stores the driver as JobProvisioningData.gpu_driver (GpuDriverInfo {vendor, version}, no DB migration): captured by the instance check for VM-based backends (this also backfills fleets provisioned before the upgrade) and at deploy time for SSH fleets.
• Surfaced as Instance.gpu_driver in the API and a DRIVER column in dstack fleet -v.
• All new fields are optional, so old/new server, CLI, and shim combinations degrade to None.
• Container-based backends stay None for now; kubernetes and vastai computes carry TODOs with implementation notes. Intel is intentionally not detected.

🤖 Generated with Claude Code