chore(chart): keda + keda-add-ons-http crds.install=false (CRDs bootstrapped per-cluster)

values.yaml

@@ -315,6 +315,16 @@ external-secrets:
 # Subchart values pass through under the dep name (`keda:`) below.
 keda:
   enabled: true
+  # CRDs are pre-installed out-of-band on each cluster (kubectl
+  # apply --server-side from the chart's templates/crds/*.yaml). The
+  # in-chart helm install path is broken on ArgoCD: scaledjobs.keda.sh
+  # has a 581 KB schema that exceeds K8s' 262144-byte annotation
+  # limit when applied client-side. Server-side apply works at the
+  # CRD level but not when ArgoCD goes through its template-and-apply
+  # pipeline. The bootstrap step on cluster connect handles install.
+  # See docs/AI_STUDIO_ARCHITECTURE.md (TBD) for the full flow.
+  crds:
+    install: false
   # operator + adapter + webhook — keep CPU/RAM modest. KEDA polls
   # event sources every pollingInterval (default 30s); on a cluster
   # with no ScaledObjects it does no work.
@@ -360,6 +370,13 @@ kedaHttpAddon:
 # buffers each cold-start request until the target pod is Ready.
 # The scaler is the control loop watching HTTPScaledObject status.
 keda-add-ons-http:
+  # HTTP add-on CRD (HTTPScaledObject) also pre-installed out-of-band
+  # for symmetry with the KEDA core CRDs above. The HTTP add-on chart
+  # itself doesn't have the annotation-size issue (its single CRD is
+  # small), but disabling chart-managed install keeps the operational
+  # contract uniform: 'CRDs are bootstrap, controllers are chart'.
+  crds:
+    install: false
   interceptor:
     replicas:
       # Scale the interceptor itself with HPA on its own metrics —