prometheus: Add Prometheus exporter and example Grafana dashboard

[sbates130272]

• Add prometheus/lemonade-exporter.py: Prometheus exporter that scrapes Lemonade Server metrics from /api/v1/stats and /api/v1/health endpoints
• Add prometheus/grafana/grafana-dashboard.json: Example Grafana dashboard for monitoring CPU, GPU, and Lemonade Server metrics
• Add prometheus/lemonade-exporter.service: systemd service file for running the lemonade exporter as a systemd service.
• Add README.md: Documentation for exporter setup and usage
• Update llamacpp_server.cpp: Add --metrics flag to enable a llama.cpp Prometheus metrics endpoint for backend monitoring
• Add decode_token_times tracking to streaming_proxy for accurate per-token metrics instead of estimating from average TPS

The exporter exposes metrics for:

• Server status and availability
• Token generation performance (tokens/sec, TTFT)
• Per-token decode timing (histogram)
• Model loading and management
• llama.cpp backend metrics (throughput, requests, decode calls)
• Cache hit rates and token processing rates

The Grafana dashboard visualizes:

• GPU utilization, VRAM, temperature, power consumption
• CPU load, memory, disk and network I/O
• Lemonade Server performance metrics
• Backend-specific metrics per model
• GPU kernel dispatch rate and workload patterns

This provides comprehensive monitoring of the Lemonade Server with an example of potential Grafana dashboard visualization.