obs(redis): lua pool metrics — r1 review fixes

Round-1 fixes on commit f7354b4:

1. coderabbit CRITICAL @ JSON:1749 — row ID 24 collision

   The new "Lua VM Pool" row reused id:24, which the existing
   "Hot Path (legacy PR #560)" row also uses for one of its
   nested panels. Grafana requires unique IDs across the entire
   dashboard; collisions cause import failures / ambiguous UI
   refs. Changed the row's id to 33 (max existing was 32). Codex
   P2 raised the same finding.

2. coderabbit MAJOR @ JSON:1817 — missing node_id template filter

   All four new Lua VM Pool queries hardcoded {job="elastickv"}
   and ignored the $node_id template variable that every other
   panel in this dashboard threads through. Operators picking a
   specific node in the dropdown got cluster-wide data on this
   row only. Added node_id=~"$node_id" to all 8 query strings
   (idle, max_idle, hits, misses, drops_rate, drops_range, and
   the saturation ratio's numerator + denominator).

3. gemini medium @ lua_pool.go:86 — AlreadyRegisteredError absorption

   RegisterLuaPool now wraps the loop with errors.As(err,
   *prometheus.AlreadyRegisteredError) and skips. Test harnesses
   that share a registry across sub-tests (and any hypothetical
   hot-reload path in main.go) can now retry without tripping a
   spurious slog.Warn. Other registration errors (invalid metric
   name, label conflict) still propagate.

Test:
  go test -race -count=1 -run TestRegisterLuaPool ./monitoring
                                                      -- green.
  golangci-lint run --config=.golangci.yaml ./monitoring ./adapter .
                                                      -- 0 issues.
  python3 -c "import json; json.load(open(
    'monitoring/grafana/dashboards/elastickv-redis-summary.json'))"
                                                      -- JSON valid.

New test TestRegisterLuaPool_IdempotentOnDoubleRegister locks in
the idempotent-double-register contract per CLAUDE.md's
test-with-fix convention.

Author: bootjp

monitoring/lua_pool.go

@@ -81,6 +81,15 @@ func RegisterLuaPool(registerer prometheus.Registerer, src LuaPoolSource) error
 	}
 	for _, c := range collectors {
 		if err := registerer.Register(c); err != nil {
+			// Idempotent on AlreadyRegistered so callers that retry
+			// (test harnesses with shared registries, hypothetical
+			// hot-reload paths) don't trip a spurious slog.Warn in
+			// main.go. Any other error — invalid metric name,
+			// inconsistent labels, registry conflict — surfaces.
+			var already prometheus.AlreadyRegisteredError
+			if errors.As(err, &already) {
+				continue
+			}
 			return errors.Wrap(err, "register lua pool collector")
 		}
 	}

monitoring/lua_pool_test.go

@@ -100,6 +100,25 @@ func TestRegisterLuaPool_NilSafe(t *testing.T) {
 	require.NoError(t, RegisterLuaPool(nil, nil))
 }
 
+// TestRegisterLuaPool_IdempotentOnDoubleRegister locks in the
+// AlreadyRegisteredError absorption added per gemini r1 review:
+// calling RegisterLuaPool a second time against the same registerer
+// must succeed silently rather than returning a noisy error that
+// main.go would surface as a slog.Warn. Test harnesses that share a
+// registry across sub-tests rely on this.
+func TestRegisterLuaPool_IdempotentOnDoubleRegister(t *testing.T) {
+	t.Parallel()
+	reg := prometheus.NewRegistry()
+	src := &fakeLuaPool{maxIdle: 64}
+	require.NoError(t, RegisterLuaPool(reg, src))
+	// Second call must be a no-op success: each of the five
+	// CounterFunc / GaugeFunc collectors AlreadyRegisteredError-
+	// returns on Register(), and the loop should absorb that.
+	require.NoError(t, RegisterLuaPool(reg, src),
+		"double registration must be idempotent so noisy retry paths "+
+			"don't crash main.go via the slog.Warn return")
+}
+
 // srcCollectorByName is a tiny helper that fetches a registered
 // collector by metric name so testutil.ToFloat64 can read its
 // scrape value. It hides the gather + scan boilerplate from each