📊 Add basic metrics endpoint

Description

Create a /metrics endpoint that exposes basic server metrics in a simple JSON format.

🔄 Current Status

Partially Implemented: The /model/info endpoint currently provides comprehensive statistics, but a dedicated /metrics endpoint would be more appropriate for monitoring systems.

✅ Already Implemented

• Metrics tracking in existing request handlers
• Thread-safe metrics collection using Arc<Mutex<ModelStats>>
• JSON format responses
• Integration tests for metrics via /model/info

Current metrics available via /model/info:

• Total inference count
• Total inference time
• Average inference time
• Error count and success count
• Memory usage estimation
• Min/max inference times
• Last inference timestamp

🔄 Remaining Work

• Add dedicated GET /metrics endpoint to src/api.rs
• Separate server-level metrics from model-specific metrics
• Add total requests counter (not just inference requests)
• Add average response time for all endpoints
• Consider Prometheus-compatible format option

Implementation Guidance

• Extract metrics logic from ModelStats to a separate ServerMetrics struct
• Keep existing /model/info for model-specific statistics
• New /metrics for server-wide metrics
• Use atomic counters for high-performance metrics collection

Estimated Difficulty

Easy - 2-3 hours (reduced due to existing metrics infrastructure)