gpu: Update runtimeClasses for correct podoverhead

We cannot only rely only on default_cpu and default_memory in the
config, default is 1 and 2Gi but we need some overhead for QEMU and
the other related binaries running as the pod overhead. Especially
when QEMU is hot-plugging GPUs, CPUs, and memory it can consume more
memory.

Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>

tools/packaging/kata-deploy/runtimeclasses/kata-qemu-nvidia-gpu-snp.yaml

@@ -6,8 +6,8 @@ metadata:
 handler: kata-qemu-nvidia-gpu-snp
 overhead:
     podFixed:
-        memory: "2048Mi"
-        cpu: "1.0"
+        memory: "4096Mi"
+        cpu: "1"
 scheduling:
   nodeSelector:
     katacontainers.io/kata-runtime: "true"

tools/packaging/kata-deploy/runtimeclasses/kata-qemu-nvidia-gpu-tdx.yaml

@@ -6,8 +6,8 @@ metadata:
 handler: kata-qemu-nvidia-gpu-tdx
 overhead:
     podFixed:
-        memory: "2048Mi"
-        cpu: "1.0"
+        memory: "4096Mi"
+        cpu: "1"
 scheduling:
   nodeSelector:
     katacontainers.io/kata-runtime: "true"

tools/packaging/kata-deploy/runtimeclasses/kata-qemu-nvidia-gpu.yaml

@@ -6,8 +6,8 @@ metadata:
 handler: kata-qemu-nvidia-gpu
 overhead:
     podFixed:
-        memory: "160Mi"
-        cpu: "250m"
+        memory: "4096Mi"
+        cpu: "1"
 scheduling:
   nodeSelector:
     katacontainers.io/kata-runtime: "true"