e2e: add provider switch (upstream/azure), bump keda-kaito-scaler to v0.5.1, instrument timings

[rambohe-ch]

1. Provider switch

• New E2E_PROVIDER variable in versions.env (default: upstream).
• 'upstream': everything via Helm/upstream manifests (current behavior).
• 'azure': enable AKS managed add-ons at cluster create time (--enable-keda, --enable-gateway-api). install-components.sh skips the Helm KEDA install and the upstream Gateway API standard-install.yaml apply, falling back only if the managed CRDs/operator are absent.
• Threaded through hack/e2e/scripts/{run-e2e-local,setup-cluster, install-components,validate-components}.sh, .github/actions/ e2e-base-setup/action.yaml, and both .github/workflows/e2e*.yaml (workflow_dispatch input 'provider').
• Derived KEDA_NAMESPACE: 'keda' for upstream, 'kube-system' for azure (managed KEDA add-on lives in kube-system; keda-kaito-scaler is installed in the same namespace so KEDA can resolve its ClusterTriggerAuthentication Secrets).
• test/e2e/cases.go: kedaScalerNamespace() + init() rewrite of the scaling case's NetworkPolicyAllowedNamespaces so 'kube-system' replaces 'keda' under the azure provider.

2. setup-cluster.sh azure prerequisites

• Idempotently install/update the aks-preview Azure CLI extension.
• Register Microsoft.ContainerService/ManagedGatewayAPIPreview feature flag and wait for it to reach 'Registered' (required by --enable-gateway-api).

3. keda-kaito-scaler v0.4.1 -> v0.5.1 in versions.env.

4. Install ordering / phasing

• keda-kaito-scaler moved from phase2 to phase1 (no install-time dep on KEDA; only emits ScaledObjects later).
• gpu-node-mocker moved from phase2 to phase1; phase2 now contains only Istio (renamed phase2-istio).

5. gpu-node-mocker CRD readiness gate

• cmd/gpu-node-mocker/main.go: new checkRequiredCRDs() runs a discovery query for karpenter.sh/v1 'nodeclaims' before NewManager. If the CRD is not yet served the process exits non-zero so kubelet CrashLoopBackOff retries until the KAITO operator finishes installing it. This is what makes the parallel phase1 ordering safe.
• install-components.sh: gpu-node-mocker rollout-status timeout bumped 120s -> 420s to absorb the kubelet backoff window.

6. Timing instrumentation

• install-components.sh: run_phase records per-task and per-phase wall-clock; a final summary table prints every phase total, the sum, and every phase/task entry so the longest task in each parallel phase is visible.
• run-e2e-local.sh: every top-level step (setup, build-push, install, validate, test, teardown) wrapped by time_step; a master summary is printed once via the cleanup trap (or directly for non-'all' runs).

7. Change test region from Sweden Central to Australia East because of quota limitations.

8 Disable NetworkPolicy tests in e2e tests to avoid blocking the other e2e tests. @tnsimon will improve networkpolicy tests lately.

Local usage:
E2E_PROVIDER=azure make e2e-up # AKS managed KEDA + Gateway API
E2E_PROVIDER=upstream make e2e-up # default, everything via Helm

Reason for Change:

Requirements

• added unit tests and e2e tests (if applicable).

Issue Fixed:

Notes for Reviewers: