Prometheus metrics endpoint (optional)

[pmady]

Add an optional Prometheus /metrics endpoint so gpu-mcp-server can serve both MCP agents and traditional monitoring.

Requirements:

• Add --metrics-port flag (disabled by default, e.g. --metrics-port=9090)
• Expose standard Prometheus metrics: gpu_utilization_percent, gpu_memory_used_bytes, gpu_temperature_celsius, gpu_power_draw_watts
• Labels: gpu_index, gpu_uuid, gpu_name
• Use prometheus/client_golang for the metrics registry
• Run metrics server in a separate goroutine

Why this matters:

• Bridge between MCP (agents) and Prometheus (dashboards/alerts)
• Single binary for both use cases
• Familiar pattern for infrastructure teams

Difficulty: Medium