Use NVML directly and cache nvidia-smi calls

[Qubitium]

Summary

Eliminates repeated nvidia-smi process spawns by adding a zero-dependency NVML ctypes binding and a TTL/LRU cache for nvidia-smi fallback. Caching is now disabled by default and exposed through Device.config.

Changes

• device_smi/nvidia.py loads libnvidia-ml.so.1 / nvml.dll / libnvidia-ml.dylib via ctypes.CDLL and binds a minimal NVML API (nvmlInit_v2, nvmlDeviceGetCount_v2, nvmlDeviceGetHandleByIndex_v2, nvmlDeviceGetName, nvmlDeviceGetPciInfo_v3, nvmlDeviceGetMemoryInfo, nvmlDeviceGetUtilizationRates, etc.).
• NvidiaDevice initializes from NVML first and falls back to nvidia-smi if NVML is absent or returns an error. metrics() also prefers NVML and only falls back to nvidia-smi.
• nvidia-smi calls go through _run_nvidia_smi and a thread-safe _NvidiaSmiCache. All cache mutations are protected by threading.Lock, so the cache is safe under free-threaded (GIL=0) builds. A TTL of 0 disables the cache; a positive value enables stale-within-TTL caching and LRU eviction.
• device_smi/config.py exposes nvidia_smi_cache_ttl (default 0.0) and nvidia_smi_cache_maxsize (default 16), available as device_smi.config and Device.config.
• Added regression tests for config validation, disabled caching at TTL=0, TTL caching, LRU eviction, thread safety, and the cached nvidia-smi path.
• Updated README.md with the new Device.config API and the default-disabled caching behavior.

Usage

from device_smi import Device

# Enable a 2-second nvidia-smi cache (disabled by default, TTL=0)
Device.config.nvidia_smi_cache_ttl = 2.0
Device.config.nvidia_smi_cache_maxsize = 32

No new runtime dependencies are introduced.

Link to Devin session: https://app.devin.ai/sessions/35c36bf93a05479a9d150d6c3b1f10d5
Requested by: @Qubitium

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring