feat(metrics): add named felis_* Prometheus collectors

Introduce internal/metrics exposing the four metric families spec §23
mandates at minimum: felis_servers_total (gauge by desired state),
felis_start_duration_seconds (histogram with Minecraft cold-start
buckets), felis_image_build_failures_total and
felis_reaper_worlds_deleted_total (counters). Collectors are
package-level vars so any subsystem records without an import cycle;
Register wires them into a prometheus.Registerer and is idempotent.

Wire registration into the operator against controller-runtime's global
Registry, so /metrics on the manager's existing metrics endpoint carries
the felis_* families. Instrument the reaper to increment
felis_reaper_worlds_deleted_total in lockstep with Summary.WorldsReaped,
at the one point a world's PVC has actually been deleted.

internal/metrics/metrics_test.go

@@ -0,0 +1,94 @@
+package metrics
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/testutil"
+)
+
+// wantNames is the exact set of metric names spec §23 mandates "at minimum".
+// Pinning them here makes a rename a deliberate, test-visible act rather than a
+// silent dashboard/alert break.
+var wantNames = []string{
+	"felis_servers_total",
+	"felis_start_duration_seconds",
+	"felis_image_build_failures_total",
+	"felis_reaper_worlds_deleted_total",
+}
+
+func TestRegisterExposesNamedFelisMetrics(t *testing.T) {
+	reg := prometheus.NewRegistry()
+	if err := Register(reg); err != nil {
+		t.Fatalf("Register: %v", err)
+	}
+
+	// Record into each collector through its typed API so a Gather emits the
+	// family — this proves the exported vars are the ones actually registered,
+	// not shadow copies.
+	ServersTotal.WithLabelValues("Running").Set(3)
+	StartDurationSeconds.Observe(12.5)
+	ImageBuildFailuresTotal.Inc()
+	ReaperWorldsDeletedTotal.Add(2)
+
+	mfs, err := reg.Gather()
+	if err != nil {
+		t.Fatalf("Gather: %v", err)
+	}
+	got := map[string]bool{}
+	for _, mf := range mfs {
+		got[mf.GetName()] = true
+	}
+	for _, name := range wantNames {
+		if !got[name] {
+			t.Errorf("mandated metric %q not exposed by the registry", name)
+		}
+	}
+	// Nothing may leak out from under the felis_ namespace — a stray name would
+	// mean a collector was declared without the namespace prefix.
+	for name := range got {
+		if !strings.HasPrefix(name, "felis_") {
+			t.Errorf("metric %q is not under the felis_ namespace", name)
+		}
+	}
+}
+
+func TestRegisterIsIdempotent(t *testing.T) {
+	reg := prometheus.NewRegistry()
+	if err := Register(reg); err != nil {
+		t.Fatalf("first Register: %v", err)
+	}
+	// A second Register into the same registry must not error: setup that runs
+	// twice (manager restart, test re-entry) should be harmless, not fatal.
+	if err := Register(reg); err != nil {
+		t.Fatalf("second Register should be idempotent, got: %v", err)
+	}
+}
+
+func TestCountersRecordExpectedValues(t *testing.T) {
+	// These collectors are package-level singletons. This test asserts on a
+	// gauge child labelled "Stopped" (no other test touches it) and a >=1 bound
+	// on the build-failure counter, so values stay stable regardless of test
+	// ordering or accumulation across the binary.
+	reg := prometheus.NewRegistry()
+	if err := Register(reg); err != nil {
+		t.Fatalf("Register: %v", err)
+	}
+
+	ImageBuildFailuresTotal.Inc()
+	if got := testutil.ToFloat64(ImageBuildFailuresTotal); got < 1 {
+		t.Errorf("image build failures = %v, want >= 1", got)
+	}
+
+	g := ServersTotal.WithLabelValues("Stopped")
+	g.Set(7)
+	if got := testutil.ToFloat64(g); got != 7 {
+		t.Errorf("servers_total{state=Stopped} = %v, want 7", got)
+	}
+
+	StartDurationSeconds.Observe(30)
+	if n := testutil.CollectAndCount(StartDurationSeconds); n != 1 {
+		t.Errorf("start_duration_seconds collected %d metrics, want 1 histogram", n)
+	}
+}