[codex] Add speculative draft Prometheus metrics

[nycdubliner]

Summary

• export speculative draft acceptance counters from the existing llama-server /metrics endpoint
• label speculative counters by spec_type using the implementation type already tracked by common_speculative_state
• document metric names, Prometheus/Grafana expressions, and curl verification

Validation

• cmake --build build-hip-rocwmma --target llama-server -j "$(nproc)"
• ran constrained Gemma 4 26B-A4B MTP server on 127.0.0.1:8084
• verified curl -s http://127.0.0.1:8084/metrics | rg "speculative|draft" before and after one chat completion; counters increased from zero to generated/accepted values
• git diff --check

Notes

The exported counters use the same in-memory source counters as the statistics <type> server log line. No log parsing or separate metrics endpoint is added.