Add Prometheus metrics endpoint; containerize integration tests

Prometheus metrics (internal/metrics/, cmd/maglevd/)
- New --metrics-addr flag (default :9091, env MAGLEV_METRICS_ADDR)
  serving /metrics via promhttp.
- Gauge metrics scraped on demand via a custom prometheus.Collector:
  maglev_backend_state, maglev_backend_health, maglev_backend_enabled,
  maglev_frontend_pool_backend_weight.
- Inline counter/histogram metrics updated per probe:
  maglev_probe_total (by backend, type, result, code),
  maglev_probe_duration_seconds (by backend, type),
  maglev_backend_transitions_total (by backend, from, to).
- StateSource interface in metrics package breaks the import cycle
  with checker; checker.Checker satisfies it via GetBackendInfo.

Integration tests
- Run maglevd inside a containerlab node (debian:trixie-slim with
  build/ bind-mounted) instead of on the host. Eliminates port
  collisions with any host maglevd.
- maglevc commands run via docker exec into the maglevd container.
- Add 6 Prometheus test cases: endpoint reachable, all backends
  report state=up, probe counters non-zero, duration histogram
  populated, pool weights correct, transition counters present.

internal/checker/checker.go

@@ -13,6 +13,7 @@ import (
 
 	"git.ipng.ch/ipng/vpp-maglev/internal/config"
 	"git.ipng.ch/ipng/vpp-maglev/internal/health"
+	"git.ipng.ch/ipng/vpp-maglev/internal/metrics"
 	"git.ipng.ch/ipng/vpp-maglev/internal/prober"
 )
 
@@ -267,6 +268,22 @@ func (c *Checker) GetBackend(name string) (BackendSnapshot, bool) {
 	return BackendSnapshot{Health: w.backend, Config: w.entry}, true
 }
 
+// GetBackendInfo returns the health state and key config fields for a backend.
+// Satisfies metrics.StateSource.
+func (c *Checker) GetBackendInfo(name string) (metrics.BackendInfo, bool) {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+	w, ok := c.workers[name]
+	if !ok {
+		return metrics.BackendInfo{}, false
+	}
+	return metrics.BackendInfo{
+		Health:  w.backend,
+		Enabled: w.entry.Enabled,
+		HCName:  w.entry.HealthCheck,
+	}, true
+}
+
 // PauseBackend pauses health checking for a backend by name. The probe
 // goroutine is cancelled so no further traffic is sent to the backend. The
 // backend's state is set to paused and remains frozen until ResumeBackend is
@@ -466,6 +483,7 @@ func (c *Checker) runProbe(ctx context.Context, name string, pos, total int) {
 			slog.Debug("probe-start", "backend", name, "type", hc.Type)
 			start := time.Now()
 			result = prober.ForType(hc.Type)(probeCtx, pcfg)
+			elapsed := time.Since(start)
 			cancel()
 			slog.Debug("probe-done",
 				"backend", name,
@@ -473,8 +491,14 @@ func (c *Checker) runProbe(ctx context.Context, name string, pos, total int) {
 				"ok", result.OK,
 				"code", result.Code,
 				"detail", result.Detail,
-				"elapsed", time.Since(start).Round(time.Millisecond).String(),
+				"elapsed", elapsed.Round(time.Millisecond).String(),
 			)
+			res := "success"
+			if !result.OK {
+				res = "failure"
+			}
+			metrics.ProbeTotal.WithLabelValues(name, hc.Type, res, result.Code).Inc()
+			metrics.ProbeDuration.WithLabelValues(name, hc.Type).Observe(elapsed.Seconds())
 		}
 
 		c.mu.Lock()
@@ -493,6 +517,7 @@ func (c *Checker) runProbe(ctx context.Context, name string, pos, total int) {
 				"code", result.Code,
 				"detail", result.Detail,
 			)
+			metrics.TransitionTotal.WithLabelValues(name, t.From.String(), t.To.String()).Inc()
 			c.emitForBackend(name, addr, t, c.cfg.Frontends)
 		}
 		c.mu.Unlock()