GPU-aware execution layer for DeepSeek Harness

Key features

• gpu_status — one query, every device: memory used/total, SM utilization, temperature, and a free/busy verdict.
• nvidia-smi ignores CUDA_VISIBLE_DEVICES — it always reports physical indices.
• A device is busy at or above 80% memory used or 50% utilization (both configurable).
• gpu_exec — one-shot command with a selected card: CUDA_VISIBLE_DEVICES= is passed through the mounted ctx.shell executor's environment.
• select count cards for multi-GPU commands.