v2.1.0: Circuit breaker, metrics, event-driven architecture

[Will-Luck]

Summary

• Circuit breaker & restart policy: Exponential backoff, per-container restart budgets, and action labels (autoheal.action = restart/stop/notify/none) to prevent restart storms
• Prometheus metrics: /metrics endpoint with 9 metric definitions — restart counters, skip counters, notification tracking, event processing histograms, unhealthy/circuit gauges
• Event-driven architecture: Docker event stream watcher with auto-reconnect and debouncing, replacing polling for instant unhealthy detection (polling fallback preserved for tests)
• Notification improvements: Rate limiting (configurable window per container), retry with exponential backoff (3 attempts for action events), per-service metrics
• Testability: Extracted docker.API, clock.Clock, notify.Notifier interfaces with mock implementations. 25+ unit tests achieving 60% coverage on guardian package
• Bug fixes: WaitGroup goroutine tracking for notifications, Warn-level error logging, empty container names guard, config validation
• CI hardening: Coverage reporting, govulncheck, Trivy container scanning, 3 new acceptance tests (opt-out, circuit-breaker, custom-label), timeout limits, failure artifact capture

New Environment Variables

Variable	Default	Description
AUTOHEAL_BACKOFF_MULTIPLIER	2	Backoff multiplier for restart delays
AUTOHEAL_BACKOFF_MAX	300	Maximum backoff delay (seconds)
AUTOHEAL_BACKOFF_RESET_AFTER	600	Healthy duration before backoff resets
AUTOHEAL_RESTART_BUDGET	5	Max restarts per rolling window (0=unlimited)
AUTOHEAL_RESTART_WINDOW	300	Rolling window for restart budget (seconds)
METRICS_PORT	0	Prometheus metrics port (0=disabled)
NOTIFY_RATE_LIMIT	60	Min seconds between notifications per container

Test plan

• go build ./cmd/guardian compiles
• golangci-lint run ./... lint clean
• go test -count=1 ./... all unit tests pass
• docker build image builds successfully
• Existing 6 acceptance test suites pass in CI
• New 3 acceptance tests (opt-out, circuit-breaker, custom-label) pass in CI
• govulncheck and Trivy scans run in CI

🤖 Generated with Claude Code