CI: switch e2e tests to large GitHub runner (#661)

Switch e2e tests workflow from `small` runners (20 cores / RAM 64 GiB)
to `large` runners (32 cores / RAM 128 GiB) to speed up the workflow.

The PR also contains a fix for `autoscaling` e2e test, which started to fail  
only after switching to `large` runners. After VM creation with 0.5 CU and 
min 0.25 CU  we check that it has 0.5 CU. It seems, on `small` runners, 
load avg is higher when we check CU, and autoscaler doesn't scale down 
the VM from the beginning as quickly as it does on `large` runners.

Author: bayandin

.github/workflows/e2e-test.yaml

@@ -25,7 +25,7 @@ jobs:
         cluster:
           - k3d
           - kind
-    runs-on: [ self-hosted, gen3, small ]
+    runs-on: [ self-hosted, gen3, large ]
     steps:
       - uses: actions/checkout@v3
         with:

tests/e2e/autoscaling/00-assert.yaml

@@ -11,7 +11,5 @@ status:
   conditions:
     - type: Available
       status: "True"
-  # This is the "canonical form" for Kubernetes. Even if we pass 1, it will work but the created resource will have "1".
-  # This assert checks Kubernetes output (not input), so we need quotes
-  cpus: 500m
+  cpus: 250m
   memorySize: 1Gi