Decouple the TGI PodMonitoring from HPA

Today, we only deploy a PodMonitoring object if HPA is enabled to
autoscale based on TGI metrics. But, TGI metrics are useful for
Observability anyway, so let's decouple this from the HPA deployment.

benchmarks/inference-server/text-generation-inference/main.tf

@@ -84,7 +84,6 @@ resource "kubernetes_manifest" "hpa-cpu" {
 }
 
 resource "kubernetes_manifest" "tgi-pod-monitoring" {
-  count = local.custom_metrics_enabled ? 1 : 0
   manifest = yamldecode(templatefile(local.tgi_podmonitoring, {
     namespace = var.namespace
   }))