fix(install): KEDA is installed after the WVA controller, forcing a restart on every kind install #1534

Problem

main() in deploy/lib/install_core.sh:69-89 runs deploy_wva_controller before
deploy_scaler_backend, so the controller boots with no scaledobjects.keda.sh CRD and
wires ScaledObject support off at startup. The watcher added in #1509 detects the CRD a
minute later and restarts the controller. Correct end state, but every
make deploy-e2e-infra pays an avoidable restart and a routine restart in install
output makes a real crash-loop harder to spot.

Scope: kind only. deploy_keda (scaler_runtime.sh:9) returns early for
openshift, and for kubernetes unless KEDA_HELM_INSTALL=true. Neither guard matches
the Makefile default ENVIRONMENT ?= kind-emulator, so that path always Helm-installs
KEDA. ./deploy/install.sh run directly defaults to kubernetes and is unaffected.

LWS already does this correctly — infra_wva.sh:246 installs it with --wait in
prerequisites, before the controller starts.

Fix

Move deploy_scaler_backend ahead of deploy_wva_controller. deploy_keda needs only
helm and the Prometheus Operator CRDs that deploy_monitoring_stack installs earlier, so
this looks safe — worth confirming there's no ordering intent I've missed.

Related: #1461 / #1509.