Add a metrics tracking guide for session quality

[YuhaoLin2005]

What

Create docs/metrics-tracking.md — a guide for measuring and tracking AI assistant session quality over time.

Should cover

1. What to measure: rule execution rate, output freshness (5-library mtime), session cost tier, rationalization frequency
2. How to measure: file timestamps, regex markers, cumulative JSON logging (all mechanical, no AI inference)
3. How to read trends: what a downward trend in rule execution actually means vs noise
4. Integration with delivery-gate: how the metrics feed into the dual-layer mechanical gate

Context

This fills the gap between "here's what to check" (delivery-gate) and "here's how the methodology works" (docs/hybrid-gate-architecture.md). A practical guide for someone who installed delivery-gate and wants to understand their metrics.

Notes

• Target audience: someone who installed delivery-gate and ran it for 10+ sessions
• Keep it concrete — show real metric examples, not abstract formulas
• See hybrid-gate-architecture.md for architectural context