Let bgp_peer_count drop to 0 (closes #92)

SetBGPPeerCount stored a count only when it was above 0, so once the
last peer's session ended, /api/v1/stats kept reporting the previous
count. It now stores every count from 0 up; a negative count, or one
too large for the int32 it is kept in, is still not stored.

Model: opus-5-5

internal/metrics/metrics.go

@@ -168,7 +168,7 @@ func (t *Tracker) RecordWithdrawal() {
 // SetBGPPeerCount updates the current BGP peer count
 func (t *Tracker) SetBGPPeerCount(count int) {
 	// BGP peer count is always small (< 1000), so int32 is safe
-	if count > 0 && count < 1<<31 {
+	if count >= 0 && count < 1<<31 {
 		t.bgpPeerCount.Store(int32(count))
 	}
 }

internal/metrics/metrics_test.go

@@ -25,3 +25,18 @@ func TestReconnectCount(t *testing.T) {
 		t.Fatalf("reconnect count = %d, want %d", got, wantReconnects)
 	}
 }
+
+// TestBGPPeerCountDropsToZero sets a peer count of 1 and then 0, the way the
+// streamer does when the last peer's session ends, and expects 0 back.
+func TestBGPPeerCountDropsToZero(t *testing.T) {
+	t.Parallel()
+
+	tracker := metrics.New()
+
+	tracker.SetBGPPeerCount(1)
+	tracker.SetBGPPeerCount(0)
+
+	if got := tracker.GetBGPPeerCount(); got != 0 {
+		t.Fatalf("peer count = %d after setting 0, want 0", got)
+	}
+}