chore(ci): remove redundant DRA test steps from inference workflow

The secure-accelerator-access conformance check is now self-contained:
it creates its own DRA test resources (namespace, ResourceClaim, Pod),
validates DRA access patterns, and cleans up automatically.

Remove the separate "Deploy DRA GPU test" and "DRA GPU test cleanup"
CI steps that previously applied docs/conformance/cncf/manifests/
dra-gpu-test.yaml as a prerequisite. The static manifest is retained
for documentation but no longer needed in the CI pipeline.

Verified: CI run 22286049313 shows TestSecureAcceleratorAccess PASS
(3.61s) in the "Validate cluster" step, confirming the self-contained
check works correctly on GPU hardware.

Author: dims

.github/workflows/gpu-h100-inference-test.yaml

@@ -31,7 +31,6 @@ on:
       - '.github/actions/load-versions/**'
       - 'tests/manifests/**'
       - 'tests/chainsaw/ai-conformance/**'
-      - 'docs/conformance/cncf/**'
       - 'recipes/components/dynamo-platform/**'
       - 'recipes/components/prometheus-adapter/**'
       - 'recipes/overlays/kind.yaml'
@@ -109,38 +108,14 @@ jobs:
           fi
           echo "Snapshot correctly detected ${GPU_COUNT}x ${GPU_MODEL}"
 
-      # --- Deploy DRA test pod (prerequisite for secure-accelerator-access check) ---
-
-      - name: Deploy DRA GPU test
-        run: |
-          kubectl --context="kind-${KIND_CLUSTER_NAME}" apply \
-            -f docs/conformance/cncf/manifests/dra-gpu-test.yaml
-
-          echo "Waiting for DRA GPU test pod to complete..."
-          if kubectl --context="kind-${KIND_CLUSTER_NAME}" -n dra-test \
-            wait --for=jsonpath='{.status.phase}'=Succeeded pod/dra-gpu-test --timeout=120s; then
-            echo "DRA GPU allocation test passed."
-          else
-            echo "::error::DRA GPU test pod did not succeed"
-            kubectl --context="kind-${KIND_CLUSTER_NAME}" -n dra-test \
-              logs pod/dra-gpu-test 2>/dev/null || true
-            kubectl --context="kind-${KIND_CLUSTER_NAME}" -n dra-test \
-              get pod/dra-gpu-test -o yaml 2>/dev/null || true
-            exit 1
-          fi
-
-          echo "=== DRA GPU test logs ==="
-          kubectl --context="kind-${KIND_CLUSTER_NAME}" -n dra-test \
-            logs pod/dra-gpu-test
-
       # --- Install Karpenter before validation so cluster-autoscaling check passes ---
 
       - name: Install Karpenter + KWOK (setup)
         run: bash kwok/scripts/validate-cluster-autoscaling.sh --setup
 
       # --- Validate cluster (Go conformance checks run inside K8s Jobs) ---
-      # Replaces previous bash assertion steps for: inference-gateway,
-      # accelerator-metrics, pod-autoscaling, secure-accelerator-access.
+      # Includes self-contained secure-accelerator-access check (creates its own
+      # DRA test resources, validates, and cleans up automatically).
 
       - name: Validate cluster
         run: |
@@ -258,12 +233,6 @@ jobs:
       - name: Cluster Autoscaling (Karpenter + KWOK)
         run: bash kwok/scripts/validate-cluster-autoscaling.sh --exercise
 
-      - name: DRA GPU test cleanup
-        if: always()
-        run: |
-          kubectl --context="kind-${KIND_CLUSTER_NAME}" delete \
-            -f docs/conformance/cncf/manifests/dra-gpu-test.yaml --ignore-not-found 2>/dev/null || true
-
       # --- Evidence collection ---
 
       - name: Collect AI conformance evidence
@@ -337,9 +306,6 @@ jobs:
           kubectl --context="kind-${KIND_CLUSTER_NAME}" -n monitoring get pods -o wide 2>/dev/null || true
           echo "=== DRA ResourceSlices ==="
           kubectl --context="kind-${KIND_CLUSTER_NAME}" get resourceslices -o wide 2>/dev/null || true
-          echo "=== DRA test pod spec ==="
-          kubectl --context="kind-${KIND_CLUSTER_NAME}" -n dra-test \
-            get pod/dra-gpu-test -o yaml 2>/dev/null || true
           echo "=== Node status ==="
           kubectl --context="kind-${KIND_CLUSTER_NAME}" get nodes -o wide 2>/dev/null || true