fix: resolve deadlock in node_config collector causing request exhaustion

The outer goroutine per-node acquired a semaphore slot and held it while
collectNode spawned inner goroutines needing slots from the same semaphore.
With maxConc=5 and 5+ nodes, all slots were consumed by outer goroutines,
inner goroutines blocked forever, and Collect() never returned — permanently
consuming an HTTP MaxRequestsInFlight slot until the server stopped responding.

Remove the redundant outer semaphore acquire (inner goroutines already manage
their own slots) and add a 120s HTTP timeout as defense-in-depth.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

collector/node_config.go

@@ -75,8 +75,6 @@ func (c *nodeConfigCollector) Update(client *Client, ch chan<- prometheus.Metric
 		wg.Add(1)
 		go func(node string) {
 			defer wg.Done()
-			sem <- struct{}{}
-			defer func() { <-sem }()
 
 			if err := c.collectNode(client, ch, node, sem); err != nil {
 				emu.Lock()

main.go

@@ -6,6 +6,7 @@ import (
 	"net/http"
 	"os"
 	"strings"
+	"time"
 
 	"github.com/alecthomas/kingpin/v2"
 	"github.com/prometheus/client_golang/prometheus"
@@ -67,13 +68,17 @@ func main() {
 	registry.MustRegister(versioncollector.NewCollector("pve_exporter"))
 	registry.MustRegister(pveCollector)
 
-	http.Handle(*metricsPath, promhttp.HandlerFor(
+	http.Handle(*metricsPath, http.TimeoutHandler(
+		promhttp.HandlerFor(
 			registry,
 			promhttp.HandlerOpts{
 				ErrorLog:            slog.NewLogLogger(logger.Handler(), slog.LevelError),
 				ErrorHandling:       promhttp.ContinueOnError,
 				MaxRequestsInFlight: 5,
 			},
+		),
+		120*time.Second,
+		"Scrape timed out",
 	))
 
 	if *metricsPath != "/" {